sign in sign up
<<
Home , Aggregate demand, Andrew Weiss (economist), Axel Leijonhufvud, Business cycle, Charles Schultze, Classical economics

American Economic Association

What Is New-Keynesian ? Author(s): Robert J. Gordon Source: Journal of Economic Literature, Vol. 28, No. 3 (Sep., 1990), pp. 1115-1171 Published by: American Economic Association Stable URL: http://www.jstor.org/stable/2727103 . Accessed: 03/05/2011 12:56

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . http://www.jstor.org/action/showPublisher?publisherCode=aea. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is a not-for- that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information and tools to increase and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Economic Literature.

http://www.jstor.org Journal of Economic Literature Vol. XXVIII (September 1990), pp. 1115-1171

What Is New-KeynesianEconomics?

By ROBERT J. GORDON and National Bureau of Economic Research

This research was supported by the National Science Foundation. I am grateful to George Williamsfor help with the data and to Steven Allen, Laurence Ball, Olivier J. Blanchard, Timothy Bresnahan, Charles Calomiris, Dennis Carlton, Robert Chirinko, Russell Cooper, , Herschel Grossman, R. Glenn Hubbard, , John Leahy, Assar Lindbeck, N. Gregory Mankiw, ,,Dennis Snower,John Taylor,, and two anonymous referees for comments on one or more earlier drafts. I am also indebted to Michael Parkin and Edmund S. Phelps for helping to establish the etymology of the phrase "new-Keynesian."

I. Introduction ing on foundations laid in the late by Stanley Fischer (1977a) and Edmund A. Background Phelps and John Taylor (1977), a large JN THE LATE 1970s it appeared that the number of authors, young and middle- U. S. macroeconomic landscape was aged alike, in the past decade have pro- being swept by a new-classical tide, and duced an outpouring of research within that had become the Keynesian tradition that attempts to an isolated backwater. In fact there is build the microeconomic foundations of still a widespread impression that the and stickiness. The adjective best and brightest young macroeconom- new-Keynesian nicely juxtaposes this ists almost uniformly marched under the body of research with its arch-opposite, new-classical banner as the decade of the the new-classical approach.2 1980s began. 1 Yet it is now apparent that the rumors of the death of Keynesian eco- 2 label new-Keynesian should be attributed to nomics were greatly exaggerated. Build- Michael Parkin(1982), who has offered me the opin- ion that he originated the term new-Keynesian the- 1 The strongest written statement of the domi- ory, not new-Keynesian . The term nance of new-classical macroeconomics among the new-Keynesian was incorporatedinto a chap- younger generation is by : "By about ter subsection in Phelps (1985, p. 562) and "new- 1980, it was hard to find an American academic ma- Keynesian model" in a chapter title in the fourth croeconomist under the age of 40 who professed -to edition of my textbook (Gordon 1990), written in be a Keynesian. That was an astonishing intellectual 1986. One of the first uses of the label new-Keynesian turnabout in less than a decade-an intellectual revo- economics in a scholarlyarticle is by Laurence Ball, lution for sure . . . the young were recruited dispro- N. Gregory Mankiw, and David Romer (1988). The portionately into the new classical ranks. . . . By word new rather than neo to describe the recent 1980 or so, the adage 'there are no Keynesians under work in the classical tradition distinguishes it from the age of 40' was part of the folklore of the (American) what in the early postwar period economics profession" (1988, p. 278). called the of old-Keynesian 1115 1116 Journal of Economic Literature, Vol. XXVIII (September 1990)

This paper extracts the essential ele- dustries, across countries, and across his- ments of new-Keynesian economics for torical intervals. We ultimately reach the an audience of professional verdict that much of new-Keynesian the- who are not specialists in the microeco- ory does not succeed in explaining these nomic foundations of macroeconomics. facts. There is no intention to survey compre- The prologue (Parts II and III) is fol- hensively every notable paper in the lowed by the core of the paper, the criti- field, but rather to sift the literature for cal review of theoretical contributions in the most important ideas and themes. the new-Keynesian literature. The re- One commentator has asserted that the view is organized by recognizing two cen- new-Keynesian literature has provided tral distinctions, the first between price too many explanations of wage and price setting in product markets and wage set- stickiness, and so we apply tough stan- ting in labor markets, and the second be- dards to the major contributions, asking tween and real rigidity. whether they make an essential contribu- The theoretical analysis in the paper is tion to an understanding of the adjust- organized into a treatment of main ment of and . In , our themes and issues (Part IV), and discus- intent is to ask what is new and what is sions of nominal rigidity in the product convincing in the large literature that col- (Part V), real rigidity in the prod- lectively has become known as the new- uct market (Part VI), and models of labor Keynesian economics. market rigidity (Part VII), followed by a conclusion (Part VIII). B. Main Themes The task of new-Keynesian economics Like its precursor a decade ago (R. is to explain why changes in the aggre- Gordon 1981), this paper differs from gate are sticky, that is, why conventional surveys not just in its intent price changes do not mimic changes in to sift and criticize rather than to provide nominal GNP. Sticky prices imply that a broad and evenhanded overview. It also real GNP is not an object of by contains a substantial empirical prologue individual workers and firms but rather before *reaching the core material on is cast adrift as a residual. Thus new- new-Keynesian theory. The prologue Keynesian economics is about the (Parts II and III) argues that there are choices of monopolistically competitive three different dimensions of price sticki- firms that set their individual prices and ness (which we will label the inertia, accept the level of real sales as a con- rate-of-change, and level effects). A brief straint, in contrast to new-classical eco- survey of the emerging literature in the nomics in which competitive price-taking new empirical industrial , to- firms make choices about . gether with a new empirical time-series Why do changes in the aggregate price investigation of price adjustment across level fail to mimic changes in nominal time and countries, reveals the essential GNP? Two main themes emerge from fact that any satisfactory theory of price the theoretical review, (1) the reasons for adjustment must explain the variability the absence of nominal GNP indexation of price adjustment parameters across in- of individual prices, and (2) the reasons why, in the absence of such indexation, macroeconomics and classical . In individual prices fail fully to reflect turn, the word new rather than neo is used for the changes in nominal GNP. Underlying recent work in the Keynesian tradition, so that it can be properly juxtaposed to the new-classical ap- the first theme is an essential element proach. of any industrial -the role of Gordon: What Is New-Keynesian Economics? 1117 idiosyncratic elements of and de- sis on universal features of microeco- mand. Firms care about the relation of nomic structure. In our treatment, price their own price to their own marginal and wage stickiness emerges from a core cost. But because idiosyncratic shocks set of microeconomic elements that are cause their own and to timeless and placeless: a technology of evolve differently than nominal aggregate transactions, heterogeneity of and demand, firms have no reason to accept factor inputs, imperfect , im- the risk involved in indexing their price perfect information, and imperfect capi- to nominal . The ab- tal markets. Because these core elements sence of nominal GNP indexation opens remove any incentive for individual the way for of real rigidity to agents to focus on nominal demand in explain the sources of nominal price making their own price-setting decisions, stickiness. their presence supports the traditional The second theme is that, in the ab- view that Keynesian economics is funda- sence of nominal GNP indexation, mentally about the macroeconomic ex- changes in individual prices will respond ternalities of individual decisions and the to changes in individual marginal costs, coordination failure inherent in a free- not changes in nominal GNP. Thus the . aggregate price level will be sticky unless C. The Dichotomy Between and firms expect changes in their own mar- Demand ginal costs to mimic changes in nominal GNP. Yet they have no such expectation. With much ground to cover, there In the framework that I label the input- are many interesting topics in macroeco- output approach, each of thousands of nomics that cannot be treated here. The heterogeneous firms is enmeshed in a coverage is limited to the determinants web of intricate supplier-demander rela- of behavior, roughly, tionships. The input-output element the division of a change in nominal GNP helps to explain why firms do not simply growth between changes in prices and assume that marginal costs will move in output, and the role of wage stickiness parallel with aggregate nominal demand: (if any) in contributing to price stickiness. Most firms do not know the identity of The entire demand side of the economy all of their suppliers, their suppliers' sup- is omitted as beyond the scope of the pliers, and so on. The input-output ap- paper. In particular, we pay no attention proach places equal emphasis on the to the reasons why aggregate demand purchase-material and labor-cost com- fluctuations exhibit positive serial corre- ponents of and points to lation, nor to the respective role of mone- models of real rigidities in the labor mar- tary and nonmonetary demand distur- ket, including the and bances in causing these fluctuations, nor insider-outsider models, to help explain to the significance of changes in the be- why prices are less flexible in some in- havior of demand and velocity dustries than in others. that have occurred in the 1980s, nor to An important empirical finding in Part the merits of monetary rules, nor to the III is that prices were sticky not just in relative merits of monetary rules versus the and the postwar nominal GNP rules. These topics on the era, but before World I. This demand side can be omitted, simply be- fact casts doubt on institutional sources cause they are not at the heart of the of price and wage rigidity, for example, conflict between new-Keynesian and labor unions, and reinforces our empha- new-classical macroeconomics. 1118 Journal of Economic Literature, Vol. XXVIII (September 1990)

Omission of the demand side from the nominal GNP will affect real output, no scope of the paper leads us to skip over matter whether its source is a change in those contributions, sometimes classified the nominal or some au- as new-Keynesian, which emphasize tonomous movement of spending on con- as a source of fluctuations sumption, , pur- in demand and as a channel chases, or . Further, nominal through which the influence of monetary price stickiness opens the way for supply is transmitted (see Olivier J. shocks, for example, a change in the rela- Blanchard and Fischer 1989, pp. 478- tive , to create macroeconomic 88). We also omit any treatment of feed- that supplement the initial backs from changes in the parameters of impact on output of the by induced aggregate price stickiness to the variance demand . The microeconomic of aggregate nominal demand (see the theories surveyed in this paper apply debate between Taylor 1986; and Brad- equally to the broad question as to why ford DeLong and demand disturbances in money and au- 1986).4We take as a precedent for impos- tonomous spending, as well as supply ing a dichotomy between supply and de- shocks, cause changes in real output.6 mand, and for assuming nominal GNP to be exogenous, Robert Lucas' famous II. The Three Dimensions of Wage and paper on the international output-infla- Price Stickiness tion -off (1973), which assumed that nominal GNP was an exogenous random A. Price Stickiness in the Presence of Policy walk. In short, we are interested here in the price times output side of the A prerequisite for any theory pur- quantity equation (MV PQ), to the ex- porting to explain wage and/or price clusion of the money times velocity side. stickiness is a demonstration that the However, our focus here on nominal phenomenon of stickiness exists in real- GNP rather than money helps to clarify world data. In Part II we begin by defin- one source of frequent misunderstanding ing three different dimensions of price in this area. New-Keynesian macroeco- nomics is not limited to the question 6For convenience, this introduction concludes Affect If with some references to the many available surveys "Why Does Money Output?"5 that overlap with this paper, or that treat particular prices are sticky, then any change in issues in more detail. Fischer (1988) provides a broad- brush survey of macroeconomics, including demand, supply, and policy; while (1988) as- 3I accept David Laidler's objection in correspon- sesses the classical Keynesian debate from the per- dence that Keynesian economics is about more than spective of high- countries designing stabili- wage and price stickiness and includes a treatment zation . (1987a) provides of "how the interferes with the an extended treatment of some of the supply-side coordination of inter-temporal choices." The new- issues that concern us here, whereas Assar Lindbeck Keynesian analysis of credit rationing and other fail- (1988) provides a briefer treatment from a European ings of the monetary system is recognized as a legiti- perspective. Blanchard and Fisher (1989, chs. 8-9) mate research activity but falls outside the scope of provide a relatively technical exposition of several this paper, which is delimited by the supply-demand new-Keynesian models. At the level of specific topics dichotomy. within the general new-Keynesian rubric, surveys 4 Nevertheless, we recognize the importance of are available on labor market developments in gen- feedback from price behavior to nominal GNP for eral (Joseph Stigitz 1986; Lawrence Katz 1988), im- the econometric estimation of price adjustment coef- plicit ( 1985), effi- ficients and devote considerable emphasis in Part II ciency wage theory (Katz 1986; Andrew Weiss 1990), to the treatment of econometric that results from new-Keynesian product-market theory (Julio Rotem- such feedback. berg 1987), and the interrelations between industrial ' This is the title of the recent survey by Blanchard organization theory and macroeconomic price sticki- (1987a). ness (Dennis Carlton 1989a). Gordon: What Is New-Keynesian Economics? 1119 stickiness and distinguish between the implying in turn that the way is open to essential role of price stickiness and the consider the meaning of price stickiness. peripheral role of wage stickiness. The For instance, if the rate of change of exposition is carried out insofar as possi- prices over the cycle is always ble with a set of identities, which clarify equal to some constant fraction (ox)of the issues without imposing any theory at all. excess nominal GNP movement, then By definition, the log of nominal GNP business-cycle movements in real output (X) must be divided between the log of (q) must soak up the remaining fraction the GNP deflator (P) and the log of real (1 -ac): GNP (Q): (otxx (4) X-P+Q. (1) q = - p =(1-(x). Reserving uppercase letters for logs of levels and lowercase letters for percent- One concludes from (4) that an economy age changes per unit of time, we take with relatively sticky prices (a small (x) the time of (1) and obtain: must exhibit correspondingly large fluc- tuations in real output, as long as fluctua- x-p + q, (2) tions in nominal demand (x) are indepen- dent the stickiness (x. which states that any change in nominal of price parameter regression GNP must be divided between a change It is tempting to estimate a line to deter- in the aggregate price level and a change equation like either of (4) in real GNP. Next, we subtract from both mine the degree of price stickiness (ox). the econ- sides of (2) the long-run equilibrium or But four crucial features of natural growth rate of real GNP (q*), and omy-level effects, inertia effects, policy use a "hat" (^) to designate variables de- feedback, and supply shocks-are ig- fined net of that trend growth rate of real nored in (4) and may invalidate any inter- output: pretation of an estimated of ox as representing a structural price-stickiness x- q*--p + (q - q*); (3) coefficient. The first problem is that (4) A A ignores level or Phillips-curve effects. It --p + q. is possible for actual output to be growing This states that an excess of nominal GNP at its long-run equilibrium growth rate growth over the long-run growth rate of (i.e., x = 0) while being off its equilib- real output (x) must be accompanied by rium growth path, that is, when there some combination of inflation (p) and a is a gap between the levels of actual and deviation of real output from that same equilibrium output. The second problem long-run growth rate (q). is the possible presence of price inertia, In many and depressions as occurs when lagged variables (espe- over the course of the industrial era the cially lagged inflation) enter into the de- economy has experienced a decline in termination of current inflation. We output and that appears to defer the introduction of level and inertia have constrained employees to work effects until the next section in order to fewer hours than they wished at the cur- concentrate on the other two basic prob- rent real wage, and firms to produce less lems with (4), which concern policy feed- output than they wished at the current back and supply shocks. price. These episodes admit the possibil- The third problem is the possible pres- ity that actual output and long-run equi- ence of policy feedback from inflation to librium output are two distinct concepts, excess nominal GNP growth, as would 1120 Journal of Economic Literature, Vol. XXVIII (September 1990) occur with a policy of monetary accom- ment coefficient, the larger the ampli- modation to price changes. Such feed- tude of output fluctuations in q. Intui- back would be implied when the central tively, because in the absence of supply attempts to peg or stabilize shocks price change depends only on rates, or with a real bills doctrine in nominal demand (x), and any policy feed- which bank automatically expand back simply "blows up" price and nomi- to meet the needs of trade. The fourth nal demand change by the same propor- problem arises in the presence of autono- tion. mous supply shocks which shift the rate We cannot, however, recover the cor- of price change up and down relative to rect value of ox in the presence of supply that predicted by (4). We now consider shocks. With a z = 10 but a model in which the interaction between no (e = 0), and with the policy feedback and supply shocks be- same values of ox and 0, (5) is satisfied comes crucial in estimating the coeffi- for p = x = 20 and q = 0. If no "z" cient of price adjustment (ox)in an equa- variable is included to capture the sup- tion like (4). The subsequent results on ply-shock effect, a simple regression of coefficient bias apply to literally every p on x will recover an incorrect value of empirical study that has attempted to re- ox = 1. In general, as shown in the notes late price or output change to such en- to Table 1, a regression of p on x in a dogenous variables as nominal or real sample containing both demand and sup- GNP, the money supply, or unemploy- ply shocks will an upward biased ment. estimate of the price-adjustment parame- Consider the two-equation model: ter ox, the larger is the accommodation parameter (0) and the larger is the vari- p = (xx + z ance of supply shocks relative to demand x= Op + e, (5) shocks (u2z/U2e). The problem cannot be avoided by replacing nominal GNP where z is the supply-shock term and e change (x) by real GNP change (q) in the is the demand shock. The coefficient of first equation in (5), because this would policy feedback (0) would be positive if introduce a negative bias that works in growth in the money supply responds reverse and is larger, the smaller the ex- positively to a contemporaneous change tent of policy accommodation. in the inflation rate. Table 1 provides examples of the bias It is easy to see that in a world with that will result in estimating the price- no supply shocks (z = 0), policy accom- stickiness coefficient (ox), when excess modation makes no difference. Here we nominal GNP growth (x) or excess real relegate the algebra to the source note GNP growth (q) are used as the alterna- in Table 1 and consider a numerical ex- tive explanatory demand growth vari- ample with a 10 percent positive realiza- ables. Columns 3 and 4 show that there tion of e, a price-adjustment parameter is no bias in using x with any degree of Ox= 0.5, and a policy accommodation policy feedback or any importance of sup- parameter 0 = 1.0. Then (5) is satisfied ply shocks, as long as both do not occur with the values p = 10, x = 20, and q = together. Using q as the explanatory vari- 10. A regression of p on x for a sample able introduces a downward bias when period with no supply shocks will recover there are supply shocks, even if there is the correct value of a-, 0.5 = 10/20. De- no policy feedback. In intermediate situ- spite policy feedback, we would correctly ations, as on lines 4 and 5, estimates us- infer that the smaller the price adjust- ing alternatively x and q bracket the true Gordon: What Is New-Keynesian Economics? 1121

TABLE 1 RANGE OF ESTIMATED PRICE STICKINESS COEFFICIENTS WIIEN THE TRUE COEFFICIENT IS (X = 0.25

Estimated Value of (x Policy Relative When Regressor Is Response Importance of Line Coefficient (0) Supply Shocks (r) q

(1) (2) (3) (4) 1 0 0 0.25 0.25 2 0 0.25 0.25 0.06 3 0 1 0.25 -0.50 4 0.5 0.25 0.35 0.12 5 0.5 1 0.60 -0.15 6 1 0.25 0.40 0.25 7 1 1 0.63 0.25

Sources by column: With reference to the model in equation (5) in the text, the estimated coefficient E(o-) in column (3) is the true coefficient ((x)plus the ratio of the covariance of x with z divided by the variance of x:

E(ot) = ot + [0(1 - o0)r]/[l + 02r],

where r = r2z/l2e. When 4 (=x - p) is used as an alternative explanatory variable in the first equation of (5), instead of x, the equation estimated is p = 134, and the estimated coefficient E(13)is

E(13)= 13- {(1 - 0)[r + 13(1- 0)]}/[r + (1 - 0)2r]. The coefficient shown in column (4) is the value of (x that would be calculated on the assumption that 13is true, aL= 13/(1 + 13).

value. Using q retrieves the correct co- is more likely to be upward than down- efficient only when policy feedback is ward biased when the demand variable complete, that is, when policy fully ac- is x, thus overstating the extent of price commodates the supply shock, as in lines flexibility and tilting the conclusions 6 and 7. against the new-Keynesian view that We reach five important conclusions prices are sticky and toward the new- from this analysis. First, to the extent classical view that prices are flexible. that demand shocks have been substan- Third, in the presence of partial policy tially more important than supply shocks accommodation, equations with nominal (at least prior to the oil-shock decade of and real GNP changes (x and q) as alter- the 1970s), the degree of price stickiness native demand variables will bracket the can be measured by the coefficient on true coefficient of price stickiness. excess nominal GNP change (x) in a Fourth, any empirical study of price ad- regression equation explaining price justment should attempt to find proxies change, even if policy has partially or for the supply shocks themselves, rather wholly accommodated price changes. than allowing such shocks to remain hid- Second, in view of the widespread view den in the error term, in order to mini- of and mize these that occur in the pres- (1963) and most commentators that mon- ence of policy accommodation. Fifth, any etary policy has been accommodative study that does not control for supply (i.e., procyclical rather than countercy- shocks is likely to reach unreliable con- clical), the price-adjustment coefficient clusions regarding the extent and/or sec- 1122 Journal of Economic Literature, Vol. XXVIII (September 1990)

ular change in price stickiness. For Pt = YQt +;, (6) instance, a conclusion that prices had A where is the log ratio of actual to natu- become more sticky since World War II ral output, and we indicate explicitly the could be subject to the criticism that time subscript that previously has been prewar price adjustment coefficients are suppressed. The supply-shock term from upward biased because of some combi- (5) is included here in each subsequent nation of (a) greater prewar policy accom- price adjustment equation, in view of our modation and (b) a higher prewar vari- previous conclusion that adjustment ance of unmeasured supply shocks. coefficients will be biased unless a careful B. Where the and Price attempt is made to control for supply Inertia Fit In shocks. A third dimension of price stickiness One reason different authors disagree is serial correlation, sometimes simply on historical changes in the extent of called inertia. A frequent specification of price stickiness is that authors have fo- the postwar U.S. inflation process com- cused on different dimensions of sticki- bines the Phillips curve and inertia:8 ness. Thus far we have characterized price stickiness by a single parameter (a), Pt = XPt_I + YQt+ zt. (7) which denotes the marginal response of When the lagged inflation term is inter- the rate of price change to a change in preted as a proxy for the the excess growth rate of nominal GNP. expected rate of inflation then is called an Yet this relation between the change in (pe), (7) ex- pectational Phillips curve. Friedman's prices and the change in demand stands (1968) natural rate hypothesis (NRH) in contrast to the relation between the states that the coefficient on pe in an ex- change in prices and the level of demand, pectational Phillips curve is unity, that is, the Phillips curve, that may come first to mind in connection with price ad- Pt Pt + Qt + Zt. (8) justment. While the Phillips curve was This expression is compatible with steady originally developed (A. W. Phillips fully anticipated inflation when actual 1958) as an association between the and natural output are equal = 0) change in nominal wage rates and the (Qt and implies that inflation steadily accel- level of , it has become erates whenever the log output ratio is common to use the Phillips-curve termi- positive. nology to label any relation between the But, as originally pointed out by rate of change of nominal prices or wages Thomas Sargent (1971), the NRH does and the level of a utilization variable like not imply that the coefficient X in (7) must the unemployment rate or detrended be unity. The coefficient on Pe in (8) could output. Here we focus on detrended out- be unity, while at the same time rational put rather than unemployment and, be- agents could form their expectations of cause our interest is primarily in price inflation by applying a coefficient X below rather than wage stickiness, we write a Phillips-curve relation for price change:7 plots of output, trend output, and the unemployment rate, see R. Gordon (1990), p. 14 for the twentieth 7 The Okun'slaw relation between detrended out- century and p. 324 for 1964-88. put and the unemployment rate holds very closely 8 In practice, the first-order autoregression on Pt-I in the postwar U. S., ensuring that any conclusions in (7) is too simple to capture the dynamics in quar- developed here for the relationshipbetween inflation terly data, and higher-order autoregressive terms and detrended output carry over to the relation be- must be included in regression estimation. In annual tween inflation and the unemployment rate. For data one or two lagged inflation terms are sufficient. Gordon: What Is New-Keynesian Economics? 1123 unity to lagged inflation, if this provided While the act term may appear to drop the best possible predictor. For instance, from the sky, in fact equation (9) can be if inflation were a random walk the opti- interpreted simply as loosening the artifi- mal predictor would be X = 1, but if cial restriction in (7) that allows only the inflation were white noise, the optimal current value of the log output ratio to predictor would be X = 0. By expressing enter. The more general form (9) allows the Phillips curve in form (7) rather than both the current and one lagged value (8), we recognize that the coefficient X of the output ratio to enter as explanatory may vary in different times and places, variables, as becomes transparent when depending on the nature of the inflation we use the identity that Q^ t= - + Xt process. Further, (7) recognizes, as (8) - Pt to rewrite (9) in either of two equiva- does not, that there may be many reasons lent forms:10 for serial dependence in the inflation rate, of which expectation formation is Pt = [1/(1 -a)][Apt_ only one, and overlapping wage and price + (c- + Y)Qt - Qt-I + ztj, or, (lOa) contracts may be among the others. Pt = [1/(1 - )][pt_J Blanchard (1987b) has stressed that + (4t + YQt + zt. (lOb) there are two dimensions of price adjust- Note that either (lOa) or (lOb) reduces ment, corresponding to the two parame- to (7) when the ot parameter is set equal ters X and in (7).9 An equation like (7) y to zero.11 If both the current and one implies that shocks to nominal aggregate lagged output term matter for the rate demand cause the economy to travel of price change, as in (lOa), this implies through loops on a diagram plotting infla- tion the ratio and in (lOb) that the rate of change of prices (p) against output (Q), is related to both the rate of change an economy with low values of X and (qt) y and the level of output. The general- has "fat loops"; that is, it exhibits rela- (%t) the Phillips-curve hypothesis tively large output fluctuations and only ization of in (9) and (10) illustrates that a slow incorporation of the change in contained nominal demand growth into the rate of the same hypothesis of price adjustment can be expressed in several alternative inflation. and that the extent of price change However, in addition to the two ad- forms, in response to a change in nominal de- justment parameters in (7), we have al- mand not a single parame- ready introduced a third parameter (a-) depends on but on the three parameters X, ax, in (4) and (5), which measures the fraction ter, and 12 of current excess nominal GNP change -y. 1 (xt)taking the form of price change. How The identity in the text, QOtQt-I + Xt - Pt" are these parameters related? The con- is identical to the identity written as equation (3) nection when we add the explanatory above, in view of the fact that qt (the rate of change of detrended output) is the same as Ot - Ot-I (the variable contained in (4) to those already change in the log ratio of actual to trend output). present in (7): " Early precursors of (lOa) and (lOb), developed and originally published in 1972-73, are reprinted Pt = _Pt-i+ Oxt+ YQt+ zt. (9) in David Laidler (1975, pp. 127, 140) and differ only in assuming that 0 = 1 and that zt = 0. 12 The inclusion of both level and rate-of-change 9 Blanchard presents an equation like (7) in which effects dates back to Richard Lipsey (1960), who ag- the rate of wage change also appears, because he is gregated a model with heterogeneous micro labor interested in the speed of transmission of cost markets characterized by limited labor mobility be- changes into price changes. But the same point ap- tween markets and showed that the rate of change plies to (7), where we are interested in the division of wages would depend on both the level and rate of nominal demand changes between price changes of change of the aggregate unemployment rate. In and output changes. Lipsey's model the economy exhibits counterclock- 1124 Journal of Economic Literature, Vol. XXVIII (September 1990)

C. Where Wages Fit In lop (1938), Lorie Tarshis (1938), and (1939). Even though Keynesian economics has tradition- these authors are known for the criticism ally been more concerned with wage ri- of the countercyclical real wage assump- gidity than price rigidity. Yet our discus- tion implicit in the General Theory, re- sion to this point has made no mention sulting from its assumption of price flexi- of wages. This is fitting, because only bility combined with nominal wage price stickiness, not wage stickiness, is rigidity, it is less well known that Tarshis a necessary condition for business cycles in 1939 soon recanted and provided evi- in real output, given a particular path dence of a relatively strong negative cor- of nominal aggregate demand. There are relation between average hourly earnings no arithmetically necessary implications and total hours worked."3 Subsequently of nominal wage rigidity for the cyclical we shall examine new evidence on the behavior of output or employment, be- cyclicality of . cause sufficient flexibility in profits could D. Rate of Change or Effects allow prices to be flexible (so that p mim- ics x even if the nominal wage rate were Equations (9) and (10) imply that absolutely fixed. Yet a world of highly there may be three quite different types flexible profits with completely rigid of price stickiness, indicated respectively wages would have economic, if not arith- by a relatively high value of the X param- metic, implications. High profit volatility eter, and by relatively low values of the for any given firm would shift the firm's cxand y adjustment parameters. The role securities out along the mean-variance of the inertia parameter X is straightfor- and raise the of ward, with a higher value of X prolonging , thus creating pressure in two di- the duration of adjustment to changes in rections, toward an increase in the flexi- nominal demand, for any given values bility of wages and toward a decrease in of the a- and y parameters, and increasing the flexibility of prices, both of which the importance of overshooting and dy- would reduce the volatility of profits. In namic adjustment loops. The distinction new-Keynesian economics there is no between rate-of-change adjustment (a-) primacy to wage rigidity as contrasted and level or Phillips-curve adjustment (y) with price rigidity, and thus no presump- is clarified by examining extreme cases tion that wages are less cyclically sensi- in which one or the other is absent. tive than prices. In fact, much of the re- When there is no rate-of-change effect search of the past half decade has been (a = 0) we are back in the simple Phillips- directed toward the of curve framework in which only the level price rigidity. of output matters. For any given values The nature of cyclical flexibility in real of the X and y parameters, the accelera- wages has always played a role in discus- tion of inflation implied by an output ra- sions of Keynesian economics, dating tio of +5 percent is the same, regardless back to the debate involving John Dun- of whether the output ratio is rapidly ris- ing or rapidly falling. The opposite extreme is of more inter- wise loops in a diagramplotting wage change against est, because it has been the focus of so the level of unemployment, while an alternative model emphasizing the inertia (A = 1) effect gener- much attention in the context of high Eu- ates clockwise loops. Barro and Herschel Grossman (1976, ch. 5) derive both types of loops as special cases, as well as the condition for one or the other 13 I am grateful to Robert Chirinko for providing type of loop to dominate. me with a copy of the Tarshis (1939) note. Gordon: What Is New-Keynesian Economics? 1125

ropean unemployment in the 1980s. An (y = 0) and with X = 1 - a-, the accelera- economy lacking a level effect (-y = 0) is tion or deceleration of inflation, as well said to be characterized by hysteresis. as the change in detrended output, de- Considerable theoretical work has re- pends only on the difference between xt cently emerged to explain hysteresis and Pt-1, that is, whether or not excess phenomena, particularly in the context nominal GNP growth ratifies the inher- of the insider-outsider model of employ- ited inflation rate: ment reviewed below in Part VII. D. Whatever the theoretical explanation, Pt - Pt-i = t(x -Pt-1) + zt, and (11) the presence of hysteresis would have Qt- = (1- Xt - Pt-1) - zt. (12) profound implications for both economic doctrine and policy.14 Friedman's NRH In short, hysteresis implies that changes posits a self-correction or level effect that in both inflation and output are com- automatically stabilizes output at its equi- pletely independent of the level of de- librium value in the presence of steady trended output, and that an economy in nominal demand growth. With no level the depths of a great depression can ex- effect, the economy could settle down perience an acceleration of inflation, no at any arbitrary distance from its equilib- matter how high the level of unemploy- rium output path (with q, = 0) and experi- ment or low the level of detrended out- ence a constant rate of inflation, with no put, if excess nominal GNP growth ex- tendency for self-correction. And, if the ceeds last period's inflation rate. NRH were abandoned, it would cast sta- Empirical estimates of the general bilization policy adrift from its previous price-adjustment model in (9) and (10) mooring, the task of steering the econ- can reveal the size of the three adjust- omy toward a fixed natural rate (Qt = 0), ment parameters (X, a-, and y) in different and open to the central implication of countries and historical eras. There re- hysteresis that any level of detrended mains the issue of which alternative spec- output or rate of unemployment, no mat- ification in (9) and (10) is preferable for ter how low or high, would be consistent estimation. As argued earlier, in the with steady inflation (at a rate that de- presence of policy feedback and unmea- pends on the of both inflation and sured supply shocks, the a- adjustment unemployment). parameter is likely to be overstated when As we see below, the pattern of price nominal GNP change (x) is included as adjustment described by hysteresis is not in (9) and understated when real GNP a novel phenomenon isolated to Europe change (q) is used instead as in (lOb). in the 1980s, for the Phillips curve or This suggests that estimates based alter- "level effect" also vanished in the United natively on both forms are preferable, States, the , and Ger- because they will "bracket" the true pa- many during the . 15 A key rameter. implication of (9) is that with hysteresis We conclude from this discussion that three parameters are required to mea- 14 A valuablecompendium of papers on hysteresis, sure the degree of price stickiness: X including a fascinating introduction that traces the measuring the extent of inertia, a- mea- history of the term hysteresis in both economics and science, is Rod Cross (1988). The first use of hystere- suring the rate-of-change or hysteresis ef- sis-based models of inflation was by Phelps (1972). fect, and y measuring the level or Phil- 15 I have emphasizedthe disappearanceof the level lips-curve effect. Any attempt to measure effect in the U.S. Great Depression in several of my papers, especially R. Gordon and James Wilcox changes in the degree of stickiness over (1981, pp. 86-92) and R. Gordon (1983, pp. 93-96). time, or differences among countries, 1126 Journal of Economic Literature, Vol. XXVIII (September 1990) may be flawed if it omits any of these Mills (1927), industrial economists have three parameters from empirical testing. been aware that the responsiveness of Further, we have seen that policy accom- prices to changes in demand differs modation of supply shocks can bias coeffi- sharply across industries. The contrast cients of price adjustment, and thus any between the flexibility of the prices of adequate empirical investigation must agricultural products, and the inflexibil- make a careful attempt to control for sup- ity of the prices of complex manufactured ply shocks as well. goods, was the point of departure of Gar- Part II began by stressing the most diner Means' (1935) administered price obvious implication of the identity link- hypothesis. In the Great Depression ing changes in nominal demand, real out- every farmer knew what Table 2 shows: put, and the price level. Changes in the price level must exactly mimic changes TABLE 2 in nominal demand if business cycles in DECLINE IN PRICE AND , real output are to be avoided. Thus the VARIOUS U.S. INDUSTRIES, 1929-33 requirements for perfect price flexibility are highly restrictive: In the context of Percentage Percentage equation (10) price changes can mimic Decline Decline in changes in nominal demand only if ot = in Price Production

1, X = 0, and y = 0. Thus any combina- Agricultural implements 6 80 tion of a rate-of-change coefficient below Motor vehicles 16 80 unity, the presence of Phillips-curve Textile products 45 30 level effects (-y > 0), or the presence of 56 20 inertia effects is sufficient to generate Agricultural products 63 6 business cycles. However, Lucas (1973) showed that Phillips-curve level effects Source: Gardiner Means (1935, p. 8). could be derived in a business-cycle model in which markets clear; thus the In an economic downturn the farmer was absence of perfect price flexibility is not the victim of a highly unfavorable twist sufficient to distinguish between new- in relative prices, because the prices of classical market- models and agricultural products fell much more new-Keynesian sticky-price models. In- than those of many manufactured goods, stead, the presence of price inertia (0 > especially the agricultural implements 0) is crucial for rejecting the new-classical listed on the first line that represent one interpretation and demonstrating the ex- of the main purchased imputs in the farm istence of price stickiness. 16 sector. Within the spectrum of manufac- tured goods, crude products like textiles III. The Variety of Historical Experience tended to exhibit more price flexibility than more finished products like tractors A. Diversity of Response Across and automobiles. Industries Unfortunately, there are few empirical Since well before the publication of studies that document these differences Keynes' General Theory, for example, systematically. and James Kindahl (1970) collected prices from buy- 16 R. Gordon (1982b) shows that the Lucas (1973) ers for a large number of products, and model can be nested in a general model of price adjustment like (9) and can be rejected in the pres- these data were analyzed by Carlton ence of price inertia. (1986) to determine if there were any Gordon: What Is New-Keynesian Economics? 1127

structural relations between seller and main unchanged for years (see also Kash- buyer characteristics and the degree of yap 1990). Carlton's finding that spells price rigidity. Carlton concludes that are sometimes short and price changes there is a significant degree of price rigid- sometimes small would appear to call into ity: "It is not unusual in some industries question the theories of new-Keynesian for prices to individual buyers to remain economists based on "menu costs" of unchanged for several years" (1986, p. price changes, reviewed below in Part 638). Unfortunately, however, neither V. D. However, this apparent implication Stigler and Kindahl nor Carlton show is subject to the caveat that if demand that, as Means suggested, the degree of and/or supply shift permanently, then "complexity" of a product is related to small price adjustments can produce price rigidity. Although Carlton did try large benefits and will be observed even to measure complexity as well as other if fixed costs are large (Carlton 1989a, structural variables, he was able to find p. 932). a significant positive correlation only be- There has been remarkably little inter- tween the concentration ratio for a prod- action between new-Keynesian theory uct and the duration of its price rigidity and the evidence provided in the emerg- (i. e., the number of months a price re- ing literature of the new empirical indus- mains unchanged).'7 trial organization (NEIO) recently sur- But it is important to stress another veyed by Timothy Bresnahan (1989). The of Carlton's findings that may be of sub- overall conclusion is that there is "a great stantial importance in assessing the theo- deal of , in the sense of ries reviewed below. By no means are price-cost margins, in some concentrated all prices rigid or do they remain un- industries" (Bresnahan 1989, p. 1052). changed for substantial periods of time: One could emphasize the words a great "The fixed costs of changing price at least deal as supporting the emphasis by new- to some buyers may be small. There are Keynesian theorists on models of monop- plenty of instances where small price olistic rather than . changes occur" (Carlton 1986, p. 638). Or one could emphasize the word "some" Specifically, "there are a significant num- to point out that the world is made up ber of price changes that one would con- of both monopolistic and competitive in- sider small (i.e., less than 1 percent) for dustries. But the matter is even more most and transaction complex: One important theme of recent types." Industries where frequent price NEIO work is that pricing behavior can changes are common include plywood alternate between collusive monopolistic and nonferrous metals, and commodities behavior and price in which a with relatively long spells of rigid prices temporarily collapses, implying that a include , paper, chemicals, , given industry is characterized neither and glass. Carlton's evidence that spells by exclusively monopolistic nor competi- of price rigidity can be both short and tive behavior. long calls into question the generality of The theme of heterogeneity extends the oft-cited study by Stephen Cecchetti along other dimensions. Product differ- (1986) which provides evidence that entiation is so pervasive that "there is newsstand prices of magazines can re- almost no industry for which the that there are more than 100 products 17 I am grateful to Dennis Carlton for suggesting is untenable: without putting more struc- the wording of the last two sentences. ture on the problem, the analyst could 1128 Journal of Economic Literature, Vol. XXVIII (September 1990) need to estimate literally thousands of mimic the behavior of nominal aggregate elasticities" (Bresnahan 1989, p. 1045).18 demand. Even so basic a distinction as Heterogeneity extends to pricing behav- Arthur Okun's (1975, 1981) dichotomy ior across firms in a single industry. For between auction and markets instance, a study of airline competition rarely surfaces in new-Keynesian writ- found not only that concentration affects ing, much less in new-classical contribu- price in airline -pair markets, but also tions. And the seminal work in under- that the identity of the competitors mat- standing the coexistence of auction and ters. Carlton stresses that a given seller customer markets has been contributed can charge different prices to different by microeconomists, especially Carlton and change them at different (e.g., 1989b), who stresses that, be- times, based on "a seller's knowledge of cause of the high costs of establishing his customers and on the optimality of auction markets, "there is no incentive non-price rationing" (Carlton 1989a). In for the efficient creation of markets" a cross-section of industries, numerous (p. 7).20 dimensions of structure appear to vary together, including mass production, B. Diversity of Response Across Time large-scale facilities, unionization, capital and Space intensity, concentration, and cyclical price rigidity, all of which are more pro- Just as challenging for theorists as the nounced in the cyclically sensitive sec- diversity of responses across industries tors of the economy, particularly durable at a particular time in a particularcountry goods. 19 is the diversity of responses across time As we shall see below, new-Keynesian and countries. Much of the empirical theory has contributed relatively little to work in this area has been within the understanding these differences across context of a debate over whether prices, industries, and as yet there has been vir- wages, or both have become less flexible tually no research that attempts to test in the postwar U.S. as contrasted with theories on a diversity of industrial data. various periods before the Great Depres- We emphasize the numerous aspects of sion (among these studies are Steven Al- heterogeneity across and within indus- len 1989; Daniel Mitchell 1985; Jeffrey tries to support several themes that Sachs 1980; Schultze 1981, 1986; John emerge below, including the importance Taylor 1986; R. Gordon 1980, 1982b). In of idiosyncratic elements of product cost related work Charles Schultze and others and demand that prevent firms from as- have examined differences in response suming, as in so many simple models, coefficients over both time and space for that their costs and product demand will the U. S. and several other major indus- trialized nations (see George Alogos-

18 Further evidence on the extent of product differ- 20 Particularly striking is Carlton's example of the entiation comes in detailed studies of international costs of running the futures markets in Chicago, con- trade, showing the countries at the same of sisting of large office buildings, expensive , development both and goods with- elaborate record keeping, and the large time cost of in the same industrial categories. See Magnus the many people involved. "A significant fraction of Blomstrom, Robert Lipsey, and Lennart Ohlsson the economy of the city of Chicago is devoted to (1989). the making of markets. If a magic spell could be 19This point was suggested in a letter from Bresna- cast to make transactions costless, the Chicago econ- han, who describes these common features of cycli- omy would be devastated, at least in the short run. cally sensitive industries as "some famous coinci- This emphasizes how far from costless the making dences about industry structure." of markets really is" (1989b, p. 6). Gordon: What Is New-Keynesian Economics? 1129

koufis and Ron Smith 1989; David Backus for other countries are consistent over and Patrick Kehoe 1988; David Coe 1989; time. 21 Schultze 1981, 1986; R. Gordon 1982a, Numerous decisions must be made in 1983). the development of tests covering such It is beyond the scope of this paper a long span of history for these nations. to track all the differences in data and These include the method of detrending specification that contribute to the vari- and the development of proxy variables ety of conclusions that these studies have for the major supply shocks, a critical is- reached; that would require a separate sue in view of the likely bias in coefficient survey on this issue alone. Some of estimates when supply shocks are left un- the disagreements, particularly about measured. Another issue is the estima- changes in cyclical behavior for the U. S., tion of parameter shifts over subintervals arise because authors often do not recog- of a long historical sample period. Details nize that there are three dimensions to on the methodology and the regression price and wage rigidity, as demonstrated estimates are provided in Appendix A, in Part II. These are the degree of inertia which shows that it is desirable to con- or serial correlation (K), the rate-of- duct the estimation with slightly trans- change or hysteresis coefficient (cx), and formed versions of (9) and (lOb). This al- the level or Phillips-curve coefficient (-y). lows us to proceed directly to Tables 3 Here we provide a link between that clas- and 4, where the underlying parameters sification scheme and historical data by are unscrambled from the transformed presenting estimates of the three param- equations and presented for different eters based on price-adjustment equa- countries and historical eras. tions (9) and (10) developed above. We The estimated parameters are pro- address two issues, differences in the re- vided for changes in prices, nominal sponsiveness of prices and wages over wages, and real wages for the U.S. in U. S. history, and differences in the re- Table 3. We are interested in the nature sponsiveness of prices over the period of changes in the three price- and wage- since 1870 for five major industrial na- adjustment parameters over time, and tions (U.S., U.K., , , and also evidence on the hotly debated issue ). of the cyclical sensitivity of real wages. The empirical equations summarized Following our analysis in Part II, two es- in this paper are estimated only for nomi- timates of each parameter are provided. nal and real output data corresponding -hand element in each column to the xt and Qt variables in the theoreti- is based on an adjustment equation in cal price-adjustment equation (9). There which excess nominal GNP growth is in- is no attempt to estimate alternative ver- cluded, and this is likely to yield an up- sions for other possible nominal and real demand variables, for example, the 21As we have been reminded by Steven Allen (1989), the standard prewar series on U.S. wages money supply or unemployment. Annual are for production workers in manufacturing and output data extend back much further must be linked with a postwar series on manufactur- than unemployment data-to 1855 for ing wages, not the wage for the nonfarm private economy that is most often used in studies limited the U. K., 1870 for France, Germany, to the postwar period. Allen concludes after an ex- and the U. S., and 1885 for Japan. Wage- haustive study that differences in measurement adjustment equations are illustrated methods in either wage or output series do not change his conclusion that the cyclical sensitivity of only for the U. S., pending a careful wages was the same in the prewar and postwar peri- study to determine whether wage data ods. 1130 Journal of Economic Literature, Vol. XXVIII (September 1990)

TABLE 3 ESTIMATED PRICE AND WAGE ADJUSTMENT PARAMETERS FOR THE U.S., 1873-1987

Inertia Effect Rate-of-Change Level Effect (A) Effect (a) (y)

(1) (2) (3) Price change 1873-1914, 1923-29 0.29 (0.37) 0.17 (0.03) 0.28 (0.28) 1915-22 0.29 (0.43) 0.69 (0.43) 0.28 (0.17) 1930-53 0.25 (0.37) 0.29 (0.03) 0.09 (0.08) 1954-87 0.87 (1.01) 0.17 (0.03) 0.28 (0.28) Nominal wage change 1873-1914, 1923-29 0.46 (0.51) 0.20 (0.07) 0.43 (0.40) 1915-22 0.46 (0.44) 0.73 (0.49) 0.43 (0.22) 1930-53 0.32 (0.37) 0.44 (0.33) -0.01 (0.02) 1954-87 0.67 (0.86) 0.20 (0.07) 0.43 (0.40) Real wage change 1873-1914, 1923-29 0.17 (0.14) 0.03 (0.04) 0.15 (0.12) 1915-22 0.17 (0.01) 0.04 (0.06) 0.15 (0.05) 1930-53 0.05 (0.00) 0.15 (0.30) -0.10 (-0.06) 1954-87 -0.20 (-0.15) 0.03 (0.04) 0.15 (0.12)

Notes: Equation specifications and details are provided in Appendix A. The left parameter in each column comes from unscrambling the coefficients of equation (9') in Appendix A, the version containing excess nominal GNP growth (it) as the rate-of-change variable; the right parameter in parentheses () comes from unscrambling the coeffi- cients of equation (9"), the version containing excess real GNP growth (4t) as the rate-of-change variable.

ward biased value of the rate-of-change (cx) and level (-y) coefficients.22 The sole parameter (cx)in the presence of supply change between pre-World War I and shocks and policy feedback. The right- post-World War II was an increase in hand element in each column replaces the inertia (K) coefficient, and this in- excess nominal GNP growth with excess crease was much greater for prices than real GNP growth, and this will tend to wages. Between 1915 and 1953, how- yield a downward-biased estimate of cx. ever, there were substantial changes. The two estimates should bracket the The cx parameter rose substantially dur- true value. The parameters listed in Ta- ing World War I, while the y parameter bles 3 and 4 are allowed to change across virtually disappeared during 1930-53. time periods and are recorded when pa- When the estimated price-change pa- rameter-shift coefficients are statistically rameters are subtracted from the wage- significant, and identical parameters across time periods indicate that such 22This finding is consistent with that of Allen's careful (1989) study, which examines only wage be- shift coefficients are insignificant (for de- havior, not price behavior. Allen's specification is tails and the significance of the shift coef- similar to mine and uses both unemployment and ficients themselves, see Tables A and B data, but no nominal GNP data or supply shock proxies, and is thus subject to a bias in the in Appendix A). unemployment or output coefficients toward zero. The single most striking finding in Ta- Allen's conclusion claims that his study finds simi- ble 3 is that neither prices nor wages lar behavior prewar and postwar, but his text re- veals that he finds the same increase in the inertia were more sticky in 1954-87 than 1873- effect (coefficients on lagged inflation) as is shown 1914, as measured by the rate-of-change in Table 3. Gordon: What Is New-Keynesian Economics? 1131

TABLE 4 ESTIMATED PRICE ADJUSTMENT PARAMETERS FOR FIVE COUNTRIES, 1873-1986

Inertia Effect Rate-of-Change Level Effect (A) Effect (a) (,Y)

(1) (2) (3) U.S. 1873-1914, 1923-29 0.29 (0.37) 0.17 (0.03) 0.28 (0.28) 1915-22 0.29 (0.43) 0.69 (0.43) 0.28 (0.17) 1930-53 0.25 (0.37) 0.29 (0.03) 0.09 (0.08) 1954-87 0.87 (1.01) 0.17 (0.03) 0.28 (0.28) U.K. 1858-1914 0.24 (0.23) 0.43 (-0.06) 0.35 (0.38) 1915-22 0.24 (0.15) 0.60 (0.33) 0.35 (0.56) 1923-38 0.20 (0.18) 0.26 (0.17) 0.09 (0.07) 1960-86 0.57 (1.00) 0.43 (-0.06) 0.35 (0.38) France 1873-1913 -0.20 (-0.38) 0.47 (0.10) 0.26 (0.03) 1925-38 0.15 (0.09) 0.47 (0.26) 0.26 (0.65) 1960-86 0.55 (0.40) 0.47 (0.10) 0.26 (0.03) Germany 1873-1913 0.00 (0.30) 0.66 (-0.11) 0.21 (0.12) 1925-38 0.00 (0.00) 0.40 (0.38) 0.07 (0.07) 1960-86 0.73 (1.08) 0.33 (-0.11) 0.21 (0.12) Japan 1888-1914, 1923-38 0.15 (0.70) 0.64 (0.02) 0.39 (0.34) 1915-22 0.15 (0.18) 0.87 (0.75) 0.39 (0.09) 1960-86 0.15 (0.50) 0.64 (0.29) 0.39 (0.24)

Notes: Sample period for the U.K. begins in 1958, for Japan begins in 1888, and for the U.S. ends in 1987. Equation specifications and details are provided in Appendix A. The left parameter in each column comes from unscrambling the coefficients of equation (9') in the appendix, the version containing excess nominal GNP growth (it) as the rate- of-change variable; the right parameter in parentheses () comes from unscrambling the coefficients of equation (9"), the version containing excess real GNP growth (4t) as the rate-of-change variable. change parameters, the results before sensitivity of real wages drops to zero.23 1930 and after 1953 suggest that real Table 4 compares the results for U.S. wages have a negligible rate-of-change prices with similar price equations for the effect but a substantial procyclical level other countries. Again the most striking effect. That is, a persistent economic finding is that the ot and y parameters boom causes steady upward pressure on were the same before World War I and the real wage, and a persistent reces- after World War II in the U.K., France, sion does the reverse. However, this and Japan, with a decline in the ot coeffi- finding is subject to the qualification cient only in Germany (and one may that the manufacturing wage data used question the linking of German data over here exaggerate the cyclical sensitivity of economy-wide rates. When the 23 The respective parameter estimates for 1954-87 equations are reestimated for the postwar in an equation for the change in the real wage are, 1954-87 period alone with the fixed- with manufacturing wage data, X = -0.33, at = 0.12, -y = -0.12. With the fixed-weight nonfarm wage weight nonfarm wage index replacing the index (spliced to the employment cost index in 1975), manufacturing wage index, the cyclical the parameters are X = 0.01, at = -0.07, -y = -0.02. 1132 Journal of Economic Literature, Vol. XXVIII (September 1990)

periods when its borders were so differ- 20 ent). In every country but Japan the iner- X AGCNP 15 Deflator tia effect was much higher after World P War II than before World War I. The 10- U.K. and Japan duplicate the jump in the cxcoefficient already observed for the U.S. during the 1915-22 interval, and both the U. K. and Germany exhibit a Excess Nominal substantial decline in the y coefficient during the interwar period. \~~~~~~~~n -10-...... C. Implications a)I 1886 1889 1892 1895 1898 1901 1904 1907 1910 1913 Is the aggregate price level highly flexible, mimicking changes in excess nominal GNP growth? Or does the aggre- gate price level live a life of its own, bear- ing little relation to excess nominal GNP -10 growth and thus allowing those nominal 1-1886 1889 1892 1895 1898 1901 1904, 1960719610 1913' in f changes to create business cycles real output? The conclusion from Tables 3 and 4 is that both these statements are 4 ndthandthe ~OutputOutput Ratio,Jpn18694 true. And many in-between responses have been observed as well. At one extreme is the very high rate- of-change coefficient for Japan through- out, and for the U. S. and the U. K. during World War I and its aftermath. Figure 1 plots the 1886-1914 data for Japan and U.S., and how large is the amplitudeof shows how closely price changes track output cycles relative to nominal GNP excess nominal GNP changes. The figure growth cycles. Again basing conclusions also exhibits cycles in the log output ratio on the average of the two figures in each that are small relative to the large ampli- column, inertia effects in all countries tude of nominal GNP changes. We have were negligible before World War II. argued above that the best estimates of These results demonstrate the strong the adjustment parameters are given by diversity of aggregate price-adjustment the average of the two estimates shown behavior that has occurred across time in each column in Tables 3 and 4. On and across countries. The variety of his- the basis of these averages, it is quite torical responses of price changes to nom- apparent in Table 4 that the U. S. has inal demand changes raises questions the smallest rate-of-change parameter (a) that new-Keynesian theorists have barely and Japan the largest, both before World begun to address. Perhaps the most War I and after World War II, with the widely noted empirical test thus far de- other countries arrayed in between. The vised by new-Keynesian economists postwar U.S. also contrasts starkly with (Laurence Ball, N. Gregory Mankiw and Japan in its strong inertia effect. The top David Romer 1988) takes as its point of frame of Figure 2 shows how loose is the departure Lucas' (1973) demonstration relation between inflation and excess that the Phillips curve becomes steeper nominal GNP growth in the postwar with a higher variance of the growth rate Gordon: What Is New-Keynesian Economics? 1133

10 little or nothing to an understanding of 4_ ExcessNominal GNP the other main findings: the similar level and rate-of-change effects before World

5 - ~ ~ - War I and after World War II, despite the higher variance of nominal demand in the earlier period; the disappearance GNP Deflator of the level effect in the Depression 0 ------~~ years; the of inertia after World War II; and the differences in price flexibility among the five countries. -5. 1950 1955 1960 1965 1970 1975 1980 1985 D. Empirical Research and the Revival of Keynesian Economics Theories are often judged on their ability to explain time-series data on ag- 550Output Ratio gregate variables. This is clearly evident in the interaction of events and ideas in the past two decades. Theories have risen and fallen in acceptance in accord with the correspondence of their predictions o-5-O--10 ------...... with the evolution of actual events in the macroeconomy. To gain perspective on -101 the development of new-Keynesian eco- 1950 1955 1960 1965 1970 1975 1980 1985 nomics, we need to understand what Figure 2. Inflation, Adjusted Nominal GNP went wrong with the old-Keynesian eco- Growth, and the Output Ratio, nomics. Our emphasis here is the empiri- , 1950-89 cal failure of the Keynesian paradigm of the , and the elements that con- and of nominal demand. Ball, Mankiw, tributed to the empirical revival of the Romer show that menu-cost theory sup- Keynesian approach in the 1980s. We Lucas but also ports the correlation concentrate on empirical aspects of the that in- makes the additional prediction contest between new-classical and new- creases in the mean rate of nomi- growth Keynesian economics, and we limit the nal demand should the steepen Phillips scope of the paper by omitting any theo- with set- curve, because staggered price retical critique of either the Lucas imper- an increase in the mean inflation rate ting fect-information (Mark I) approach or the of increases the frequency price changes. real-business-cycle (Mark II) variant of in Thus far, their empirical work support new-classical macroeconomics.25 of this theoretical prediction has been subject to substantial criticism.24 In rela- 25The Lucas (1972, 1973) imperfect information tion to our empirical results of Tables 3 approach (Mark I) is now widely viewed as uncon- and 4, either the Lucas or the Ball, vincing, because it is undermined by the availability and Romer can of information on the aggregate price level and money Mankiw, approach help supply over a much shorter time period than the to explain why prices became more flexi- duration of the average . Major contri- ble during World War I but contribute butions to real-business-cycle theory (Mark II) in- clude Finn Kydland and Edward Prescott (1982) and 24 See the numerous criticisms of the paper by Ball, Prescott (1986). A generally supportive survey is pro- Mankiw, and Romer (1988) contained in the discus- vided by (1989), and critical surveys sant comments by , Andrew Rose, include Gregory Mankiw (1989) and Bennett McCal- and , as well as by Christopher Sims. lum (1989). 1134 Journal of Economic Literature, Vol. XXVIII (September 1990)

In the 1960s Keynesian economists in- fundamental weakness of the non-mar- corporated into their theoretical and ket-clearing approach. econometric models an exploitable nega- The long-run trade-off result was aban- tive long-run Phillips-curve trade-off be- doned within five years of Friedman's tween inflation and unemployment. The presidential address.26 This allowed the acceleration of inflation after 1965, to- gradual-adjustment of the gether with the positive correlation be- 1960s-style wage and price equations to tween inflation and unemployment ob- be combined with the long-run neutrality served during much of the 1970s, caused property advocated by Friedman. The ef- the mid-1960s Keynesian orthodoxy to fects of supply shocks, including the rela- unravel. In flowery language that tive prices of oil and , were ab- amounted to a simultaneous declaration sorbed into the U. S. Phillips-curve of war and announcement of victory, Lu- framework in my work of the mid-1970s, cas and Thomas Sargent (1978, pp. 49- which was developed alongside the work 50) described "the task which faces con- by David Laidler, Michael Parkin, and temporary students of the business cycle their collaborators in the open-economy [as] that of sorting through the wreckage setting of the U. K.27 The result was an . of that remarkable intellectual event econometric analogy to the dynamic ag- called the . gregate demand and supply model that It is not widely recognized that the was introduced with the 1978-79 publi- empirical reconstruction of Keynesian cation of a new generation of economic economics occurred prior to the wave of principles and intermediate macro theoretical work that is now most com- textbooks.28 Now a single reduced-form monly associated with the term new- econometric equation for price change, Keynesian economics. Lucas and Sargent like those summarized in Tables 3 and were only partly right. Yes, the predic- 4 above, could incorporate the effects of tions of the late 1960s were incorrect, gradual adjustment, of demand shocks but incorrect forecasts do not provide de that created a temporary positive correla- facto proof that a doctrine's theoretical tion between inflation and output, and underpinnings are fundamentally flawed. of supply shocks that created a temporary The essential element of Keynesian doc- negative correlation. By the end of the trine is non-market-clearing, which in 1970s the supply side of the economy turn requires the gradual adjustment of had been opened up to outside influ- prices. The 1960s version of the Phillips ences, and the list of relevant supply relation combined three elements, (1) shocks for the U. S. had grown to include gradual price adjustment, (2) a long-run not only price controls and oil shocks, trade-off, and (3) a closed-economy, de- mand-only approach with no role for im- 26 Simultaneous work by me (R. Gordon 1972) and port prices or supply shocks. Yet only by Otto Eckstein and Roger Brinner (1972) showed (1) is necessary to maintain the essence how postwar wage and price data could be made consistent with long-run neutrality. of the Keynesian paradigm, non-market- 27 My two papers were R. Gordon (1975, 1977). clearing. The other two elements, (2) and See Laidler and Michael Parkin (1975, esp. pp. 759- (3), were ephemeral empirical results, 74) for a comprehensive survey of research on wage and price equations in the late 1960s and early 1970s. based mainly on the 15 or 20 years of (1972) presents a survey of U.S. U.S. postwar data, that revealed more work on econometric price markup equations. of the short time horizon and closed- 28 As author Alan Blinder described the aggregate demand and supply model as developed in his own economy mentality of the first generation textbook, "now the Marshallian scissors come in a of econometric model builders than any giant economy size." Gordon: What Is New-Keynesian Economics? 1135

but also changes in non-oil import prices, "wreckage" in the mid-1980s, it was the exchange rates, rates, and the mini- few abortive attempts to estimate price mum wage.29 equations within the framework of Mark This so-called gradual-adjustment I new-classical macroeconomics, particu- price-change equation is completely non- larly those by (1977a, 1978; structural and as such is in principle Barro and Mark Rush 1980). So strongly highly vulnerable to the Lucas (1976) cri- was price inertia embedded in the U.S. tique. We have seen in Tables 3 and 4 data that Barro could explain price move- that coefficients of price adjustment are ments only by entering a distributed lag subject to substantial change when there of between four and six years of monetary are major changes in the economic envi- surprises that themselves only lasted a ronment, as in World War I or the Great single quarter. Why agents should be re- Depression. The sharp U.S. acting with a four-year lag to a one-quar- of the 1980s posed a formidable challenge ter monetary surprise was never ex- which the empirical price-adjustment plained. The attraction for the economics equation could have failed but did not. profession of the empirical versions of A central implication of the resuscitated Mark I new-classical macro, like the the- 1980-vintage empirical Phillips curves, oretical versions, was undermined by the the value of the sacrifice ratio of lost out- discrepancy between the time lags in- put required to achieve a permanent de- volved in data dissemination, measured celeration of inflation, turned out to be in days or weeks, as contrasted to the surprisingly close to predictions made in lags of price changes in response to nomi- advance. This suggests that, at least in nal demand shocks, measured in years the U. S., the substantial changes in or half-decades. price-adjustment parameters observed in Tables 3 and 4 to have occurred in previ- IV. New-Keynesian Theory: Common ous historical eras have been largely ab- Features sent in the postwar U.S. setting.30 A. Essential Features of Keynesian The empirical stability and predictive Economics success of the resuscitated U.S. Phillips curve is highly ironic in view of the in- The essential feature of Keynesian flammatory language used by Lucas and macroeconomics is the absence of contin- Sargent. If anything lay smoldering in uous . Thus a Keynesian model is by definition a non-market- 29 These additional supply-shock factors are omit- clearing model, one in which prices fail ted in Tables 3 and 4, as their effects are hard to to adjust rapidly enough to clear markets discern in century-long annual data samples and in- within some relatively short period of stead require the finer discrimination possible with postwar quarterly data and with the improved fixed- time. Common to almost all Keynesian weight price and wage indexes available only in the models is the prediction that in response postwar period. to a decline in nominal demand, the ag- 30 Depending on the exact used and the criterion of what constitutes a permanent slow- gregate price level will decline less than down in the inflation rate, the U.S. sacrifice ratio proportionately over a substantial time observed during the disinflation of the 1980s was period, during which the actual price between 5 and 7. An estimate of 6.2 was calculated on the basis of data through 1980 in R. Gordon and level is above the equilibrium price level King (1982, Table 5, line 3); reasons for preferring consistent with the maintenance of the this version were given in that paper (pp. 236-37). initial equilibrium level of real output. Blanchard (1984) also provides evidence from a quite different specification that the curve remained The fact that the price level is too high relatively stable during the Volcker disinflation. means that the subequilibrium level of 1136 Journal of Economic Literature, Vol. XXVIII (September 1990) output actually produced is not chosen must begin from monopolistic or imper- voluntarily by firms and workers, but fect competition, not perfect competi- rather is imposed on them as a constraint. tion, because Keynesian agents are It is the decline in nominal demand to- inherently price setters, not price gether with the absence of full price ad- takers. justmenit that causes the economic sys- A central theme of both new-classical tem itself to impose the constraint on and new-Keynesian macroeconomics is each ; nominal demand is insuffi- that accurate empirical predictions are cient to generate adequate real sales at necessary but not sufficient conditions of the actual price level. Each agent faces an acceptable theory. In addition, a the- a constraint that is indirectly a result of ory must have microeconomic founda- its own failure to reduce sufficiently its tions in the behavior of -maximiz- price and this points to a coordination ing and profit-maximizing individual failure as a central ingredient in the de- agents. The search for tractable analytic scription of Keynesian price stickiness. models to form the micro foundations of- So many people now refer to new-clas- ten leads analysts astray, causing them sical models as equilibrium business-cy- to lose sight of the forest as they construct cle models that the word equilibrium has their single exquisitely proportioned been co-opted as meaning the opposite tree. Almost all new-classical theory is of the term Keynesian. This leads some conducted in the analytically convenient commentators to label an approach that setting of " models,"> is the opposite of equilibrium economics where one can move back and forth be- as disequilibrium economics. In one tween the individual agent and the aggre- sense this is mere semantics; it does not gate economy simply by adding or re- matter whether we describe the U. S. in moving i subscripts, without having to 1932 or Europe in the mid-1980s as be- consider such analytically inconvenient ing in a of disequilibrium or low- issues as coordination failures or the employment equilibrium. However, the speed of price adjustment. Professional adjective disequilibrium is not helpful, microeconomists, as distinguished from as it conveys "a failure of agents to realize macroeconomists who dabble in micro- perceived " (to use Rob- economic modeling, find the failure to ert Barro's provocative 1979 phrase). confront aggregation seriously to be the Rather, it is best to regard the core fea- most critical flaw of representative agent ture of Keynesian economics as the grad- modeling.3' A surprising number of new- ual adjustment of prices and its corollary, Keynesian models in common the that output and employment are not neglect of aggregation; the aggregate choice variables. economy is simply the representative In contrast to new-classical equilib- agent multiplied by n. Accordingly, we rium models, with their price-taking shall find unsatisfactory those new- firms ("yeoman barbers") making volun- Keynesian models that neglect aggrega- tary choices of the output level, Keynes- tion issues, and we shall emphasize the ian non-market-clearing models turn the central role of interactions among agents, role of prices and output on their head, including coordination failures, macro- with demand-taking firms making volun- economic externalities, and producer- tary choices of the price level. Thus supplier relations. price-setting behavior is the essence of Keynesian economics. Any attempt to 31 John Pencavel suggests to me that this critical imbed it in microeconomic foundations view by microeconomists is widespread. Gordon: What Is New-Keynesian Economics? 1137

B. Micro Agents, Macro Spillovers, and all based on profit-maximizing behavior Coordination Failures at the micro level. But all this work was carried out within a The development of new-Keynesian framework, assuming in particular that economics in the past decade has primar- both real income and the price level were ily involved the search for rigorous and given. A useful distinction can be made convincing models of wage and/or price between micro theorizing at the level of stickiness based on maximizing behavior individual demand and supply functions, and . The ground and micro analysis of the market mecha- rules of this search are commonly ac- nisms (especially the ) cepted. The key ingredient in the now- whereby the actions of maximizing abandoned Mark I new-classical ap- agents are coordinated.32 Even before proach was not rational expectations, but the advent of new-, rather the assumption of continuous mar- the work of Robert Clower (1965) and ket clearing, as is evident in the labels Axel Leijonhufvud (1968) stressed inter- new-classical macroeconomics or equi- actions and spillovers among markets and librium macroeconomics. Most new- argued that the nexus of research should Keynesian models combine rational ex- shift from a partial to a general equilib- pectations with maximizing behavior at rium setting. the level of the individual agent. Any at- An interesting aspect of recent U.S. tempt to build a model based on irra- new-Keynesian research is the near-total tional behavior or submaximizing behav- lack of interest in the general equilibrium ior is viewed as cheating. No new- of non-market-clearing mod- Keynesian wants to build a model with els. That effort is viewed as having agents that Barro (1979) could criticize reached a quick dead end after the in- as failing "to realize perceived gains from sights yielded in the pioneering work of trade." So the game is to tease a failure Barro and Herschel Grossman (1971, of macro markets to clear from a starting 1976), building on the earlier contribu- point of rational expectations and the tions of (1965), Clower, and maximization of profits and individual Leijonhufvud. Explaining sticky wages at the micro level. In short, ef- and/or prices is viewed as a tough task, fects of changes in nominal aggregate de- and no one is prepared to anticipate its mand on real output and employment achievement by examining broader theo- are derived in models characterized by retical implications.33The disdain shown equilibria in which each individual agent by new-Keynesian theorists for the work takes only those actions that make him of Barro and Grossman, and the latter better off and in which no agent foregoes evolution of that line of research in the an opportunity to take advantage of a hands of Malinvaud, Muellbauer and gain from trade." Portes, Benassy, Grandmont, and oth- The recent development of microfoun- ers-notably all Europeans-is under- dations for wage and price stickiness does standable in light of the primacy of micro not, of course, represent the first attempt foundations models as the prerequisite to develop micro underpinnings for Keynesian economics. The work of 32 I am grateful to David Laidler for suggesting Friedman and on con- this distinction. sumption, Dale Jorgenson on invest- 3 In defense of the new-Keynesian approach, An- drew Weiss has suggested to me that "we have to ment, and and James solve the partial equilibrium problems first; these Tobin on the were also are the most interesting." 1138 Journal of Economic Literature, Vol. XXVIII (September 1990) for macro discourse.34 Nevertheless I ent between economic states that offer find that even the most perceptive new- relatively large and small amounts of lei- Keynesian commentators tend to forget sure. Such models stand Keynesian eco- the central message of these models. nomics on its head, because any satisfac- This message is that spillovers between tory explanation of business cycles that markets imply that the failure of one mar- warrants the label Keynesian must incor- ket to clear imposes constraints on agents porate not just price stickiness, but in in other markets. Most notably, when addition some element that explains the firms in a experience a decline evident unhappiness of the employed in in sales at the going price, this excess recessions and depressions. Further, supply of commodities "spills over" into such models fail to explain why the ad- a decline in labor demand at the going justment costs that lead to price sticki- real wage. In this light, I am sometimes ness do not in parallel imply wage sticki- surprised to read otherwise sensible ness. commentators refer to the inconsistency One important exception to this ne- of one or another new-Keynesian expla- glect of macroeconomic constraints and nation with microeconomic evidence on spillovers is the seminal work of Russell the of labor supply. Such evi- Cooper and Andrew John (1988) on mac- dence is simply irrelevant for Keynesian roeconomic coordination failures. In sev- macroeconomics. In a genuinely Keynes- eral new-classical models in which agents ian model, agents are not in a position set output, they show that spillovers and to choose the amount they work or pro- strategic complementarities can arise at duce as output varies over the cycle, and the levels of preferences and technology so the constrained amount that they do or in the organization of transactions. work or produce cannot be interpreted They reach the same conclusion as Barro as tracing movements along a choice-the- and Grossman (without making the con- oretic labor supply curve or production nection) that macroeconomic quantities function. belong in microeconomic choice func- Much existing new-Keynesian theoriz- tions. Almost alone among recent Ameri- ing is riddled with inconsistencies as a can authors in the new-Keynesian tradi- result of its neglect of constraints and tion, Cooper and John cite Jean-Pascal spillovers, and its focus on single mar- Benassy's fixed-price models (1975, 1982) kets, one at a time, in a partial equilib- and conclude for such models that "stra- rium framework. For instance, several tegic complementarity is a distinguishing of the most prominent models of price element of models with Keynesian fea- determination in the presence of adjust- tures" (1988, p. 461). ment costs limit the source of price sticki- The contribution of Cooper and John ness to the product market; they often reaffirms the traditional view (see partic- assume a perfectly competitive labor ularly Leijonhufvud 1981) that coordina- market in which workers slide up and tion failures represent the core problem down their labor supply curves, indiffer- in macroeconomics. In response to a nominal demand change, no single pri- 34Research on or fixed- vate agent has an incentive to move its price models appears to have become a specialized price exactly in proportion unless it be- European activity in macroeconomics, with near-total lieves that all other agents will do like- invisibility in a recent survey I conducted of first- year graduate macro reading lists at the top ten Amer- wise, and will do so without delay. In ican economics departments. Tobin's example, Gordon: What Is New-Keynesian Economics? 1139

No one can see the spectacle in the theater or nominal demand rather than individual stadium if everyone stands, but who has the marginal cost.35 incentive to obey a general admonition to sit down. When the teacher tells her grade school C. Real Rigidities, Nominal Rigidities, class there will be no picnic unless all gum- and the Indexation Puzzle chewing ceases, would any rational child who shares the general liking of gum stop? Threats Two central distinctions are required against everybody in general addressed to no- as a preliminary to any summary of re- body in particular rarely work. (1989, p. 15) new-Keynesian work. The first is between price setting in product markets The same point can be put differently: and wage setting in labor markets. The Rational microeconomic agents care second distinction is between nominal about the relation of their own price to rigidity and real rigidity. their own costs, not to aggregate nominal The necessary condition for non-mar- demand. Unless a single agent believes ket-clearing is a barrier to the full adjust- that the actions of all other agents will ment of nominal prices, that is, some- make its marginal costs mimic the behav- thing that prevents movements in ior of nominal demand with minimal lags, nominal prices that are equiproportion- the aggregate price level cannot mimic ate to movements in nominal demand. nominal demand, and Keynesian output However, some of the new-Keynesian fluctuations result. theories explain real rigidities as the A notable limitation of most recent for- stickiness of a wage relative to another mal models related to coordination fail- wage, of a wage relative to a price, or ures, including Cooper and John and Ste- of a price relative to another price. Expla- ven Durlauf (1989, esp. p. 110), is a nations of real rigidities in product mar- classical setting of competitive output kets include customer markets, inven- setters, rather than a Keynesian world tory models, and theories of markups of monopolistic price setters. In Dur- under , while lauf's words, those of labor markets include implicit the hallmark of this class of theories is the com- contracts, efficiency wages, and insider- patibility of different levels of real activity with outsider models. But theories of real ri- the same microeconomic specification of indi- gidities are subject to the criticism that vidual firms and consumers. The key source they do not explain nominal rigidity, be- of the multiplicity of long-run equilibriums is cause nothing prevents each individual the positive effect that high production by some set of agents has on the decision of others to agent from indexing its nominal price to produce. nominal aggregate demand. There is surprisingly little discussion This approach, based in part on seminal in recent new-Keynesian papers of opti- research by (1982, 1984), mal indexation nor of the relation be- essentially concerns the cyclical behavior tween the absence of full indexation and of productivity, the positive response of the sources of nominal rigidities. Jo Anna which is claimed to reflect "thick mar- Gray (1976), Fischer (1977b), and others kets" as a result of "positive complemen- showed in the mid-1970s that full CPI tarities." However, this has little to do indexation is not optimal in the presence with the essential Keynesian coordina- of supply shocks. Intuitively, no agent tion failure, the absence of incentives for price-setting agents to move their indi- "In related work Peter Howitt (1986) calls this vidual prices in tandem with aggregate effect a "thin market ." 1140 Journal of Economic Literature, Vol. XXVIII (September 1990)

can affordan indexed contract that rigidi- so rarely observe indexed contracts con- fies real wages and relative prices if sup- tingent on other variables (whether econ- ply disturbances continuously shift the omy-wide like nominal GNP or idiosyn- optimal . However, Gray's cratic variables like firm sales or profits). argument only supports indexation to a The primary barrier to indexation may mix of price indexes preferred by firms be the costs of making contracts more and workers, not zero indexation. Failing complicated, particularly when it is rec- to index is tantamount to linking the ognized that there are conflicts along at prices and wages of individual agents to least two dimensions. First is the Gray- a price index whose value is constant, Fischer point that the presence of aggre- and this becomes increasingly irrational gate supply shocks makes incomplete as the inflation rate increases.36 indexation optimal, and second is the Further, full indexation of the wage presence of firm-specific shocks that cre- rate to nominal GNP escapes most of the ate a conflict between the general market theoretical objections to CPI indexation, basket to which workers would prefer to because nominal GNP indexation leaves index, and the firm-specific variables to the price level free to move to equate which firms would prefer to index. Par- the real wage to the marginal product ties to a contract may differ not only in of labor. Adopting our previous notation their objective functions, but also in their with lowercase letters representing perceptions of the relative importance of growth rates, the condition necessary aggregate demand shocks, aggregate sup- for labor's share in national income ply shocks, and firm-specific shocks, and to remain constant is that the growth these perceptions may change continu- rate of the real wage (w - p) equals the ously, requiring that the form of indexa- growth rate of labor's average product tion in each new contract be negotiated (q -n): from scratch. As we review sources of rigidities in w - = q - which occurs if p n, Keynesian models, we shall return to the w = p + q - n = x - n. issue of nominal GNP indexation. Are Thus the growth rate of the nominal wage the nominal rigidities adequate to explain rate should be indexed to the growth rate the real-world absence of such indexa- of nominal GNP per unit of labor input tion? How are the two distinctions, prod- (x - n). If an adverse supply shock re- uct versus labor market and real versus duces labor's average product, then such nominal rigidities, related to each other? indexation allows the needed reduction We begin our inquiry by reviewing mod- in the real wage, whether nominal GNP els of nominal price rigidity in product growth remains constant and the rate of markets, beginning with the elementary inflation increases, or whether the infla- example of a textbook monopolist. This tion rate remains constant and the growth example implies that the response of rate of nominal GNP decelerates. price to a shift in demand is conditional Fischer (1986, pp. 152-53; 263-69) has not just on the elasticity of demand and pondered why indexation to the price the shape of the marginal cost curve, but level is so often incomplete, and why we crucially on the shift (if any) of marginal cost in response to demand. Thus prod- 36 McCallum (1986, p. 409) argues that linking to uct and labor market rigidities are com- a constant price index instead of to the CPI would be chosen only by those agents whose most preferred plementary and may be of equal impor- index is negatively correlated to the CPI. tance. Gordon: What Is New-Keynesian Economics? 1141

V. The Search for Structure: Nominal P Price Rigidity in the Product Market

A. The Textbook Monopolist and the Behavior of Marginal Cost

The behavior of a textbook monopo- list is part of relative price theory, and therefore would appear to belong in our ADo subsequent discussion of real rigidities. | \', Ip"Required However, the monopolist model has been used to derive the conditions under which costs of adjustment create a barrier MR, MR0 to changes in nominal prices. This ex- plains the connection between theories Qi Qo Q of nominal price stickiness and the tradi- Figure 3. tional partial equilibrium analysis of a price-setting monopolist illustrated in sumptions incorporated into Figure 3, Figure 3. Note that two special assump- we alter the path of the price level at tions are made in drawing Figure 3, that output levels away from the initial level the demand curves are linear and that QObut not the basic conclusion about the the marginal cost curve is horizontal. Im- required drop in marginal cost for the plications of dropping both of these as- economy to remain at Qo. For instance, sumptions are discussed shortly. replacing the special assumption of a lin- In response to a shift in the demand ear with a demand curve curve from Do to D1, quantity will change of constant elasticity, while retaining the unless there is an equiproportionate shift assumption of a horizontal MC schedule, in nominal marginal revenue and nomi- point C would lie directly to the left of nal marginal cost at the original level of point A, and the price level would remain real output. The implied marginal cost fixed in the presence of any leftward curve that maintains a constant level of movement of the demand curve. Second, output (QO) is labeled "Required" MC1. replacing the horizontal MC schedule If, following the decline in demand, mar- with a positively sloped MC schedule go- ginal cost drops instantly to the "Re- ing through point D would move points quired" MC1 line, then the intersection C and F down and to the right, thus in- of MR and MC will drop from E to G, creasing the response of the price level and the price will fall by exactly the verti- to the decline in demand and corre- cal displacement of the demand curve, spondingly reducing the output re- from POto P2. Any source of incomplete sponse. adjustment in marginal cost can then ex- This analysis suggests that the primary plain an incomplete adjustment of price. reason for sticky price adjustment is the For instance, if the marginal cost sched- sticky adjustment of marginal cost. This ule remains fixed at MCo, the intersec- would appear to place the analysis of cost tion of MR and MC occurs at point F, stickiness at the top of the new-Keynes- the new price is P1, and the new quantity ian research agenda. From the stand- is Ql. point of the aggregate economy, the most When we loosen the two special as- important cost component is labor cost, 1142 Journal of Economic Literature, Vol. XXVIII (September 1990) suggesting the familiar idea from the old rules, which call for price changes if the Keynesian economics that wage inflexi- optimal price strays outside of bound- bility is the key element in price sticki- aries determined by menu costs of price ness. However, from the standpoint of adjustment, and (2) time-dependent the individual firm, labor cost may be rules, which call for price changes at fixed less important than purchased materials and predetermined intervals written into as a component of cost, and this recogni- contracts and are in turn presumably tion elevates to the top of the research based on the costs of negotiating new agenda, along with wage determination, contracts at more frequent intervals. This the formation of expectations by individ- branch of new-Keynesian economics ual producers about the prices of pur- reinterprets these rules as profit-maxi- chased materials. mizing when menu-type or negotiation- While Figure 3 identifies the rigidity type costs of adjustment are taken of marginal cost as the key ingredient into account. Then we turn to sources in price stickiness, it also leaves open a of stickiness in marginal cost, both role for direct barriers to the adjustment in prices of purchased materials and in of price to the profit-maximizing level for wages. the monopolist, that is, to point C in the It should be clear from this analysis case of a fixed MC schedule or to point that the labor market and product market D in the case of a fully flexible MC sched- may be equally important in contributing ule. Point B represents a price above the sources of price rigidity. There has been profit-maximizing levels C or D, and some tendency to stress product markets could be explained by costs of adjustment relatively more in recent research and of the price level emphasized by new- to search for some source of nominal ri- Keynesian theorists under the general gidity for prices in the form of state- heading of menu costs or by old-Keynes- dependent or time-dependent rules. Yet ian economists who emphasized rules of it is clear from the monopolist example thumb like fixed markups of price over that any source of nominal rigidity will long-run average cost. If the price level do: A for wage adjustment will is predetermined at point B, while mar- make marginal cost sticky and indirectly ginal costs are predetermined along create a source of nominal price sticki- the schedule MCO, output and employ- ness, even if costs of adjusting prices are ment may vary up and down in re- completely absent. sponse to variation in product demand B. The Representative-Agent Model of without a change in the real product wage. This analysis of Figure 3 helps to orga- It is clear from Figure 3 that the mere nize our treatment of recent new- presence of monopolistic competition Keynesian research on the sources of does not create a presumption of price price stickiness. First we examine the stickiness. Some ingredient must be in- studies of direct barriers to price adjust- troduced either as a direct barrier to in- ment, independent of the behavior of stantaneous price adjustment or as a marginal cost, which cause the price to source of sticky marginal cost. In the re- deviate from the price that would be set cent new-Keynesian literature this point by a profit-maximizing monopolist who is most often made in the context of a has no costs of adjustment to consider. simple model of a representative-agent These direct barriers may be subdivided monopolist developed by Blanchard and into two categories, (1) state-dependent (1987) and described Gordon: What Is New-Keynesian Economics? 1143

as the canonical model of monopolistic There is no role for sticky marginal cost competition by Fischer (1988).37 There in the Blanchard and Kiyotaki "pure" are n identical producer-consumers pro- model of monopolistic competition, be- ducing goods that are imperfect substi- cause the imposition of symmetry across tutes, and there are no purchased materi- identical representative-agent producers als. Nominal aggregate demand depends has the effect of implicitly indexing both only on the nominal money supply. Mar- the relative price (Pi) and marginal cost ginal cost consists of the marginal disutil- to the aggregate price level, which in ity of production for each producer-con- turn depends only on the nominal money sumer. The canonical model describes supply. Thus the new-Keynesian theo- the determinants of output and of the rists recognize that they must go beyond desired relative price (Pi/P). With con- the mere introduction of monopolistic stant returns in production and a constant competition in order to locate the sources marginal disutility of work, the model is of price stickiness. One route is to study equivalent to Figure 3 above with a flat direct barriers to nominal price adjust- MC schedule and a constant-elasticity ment in the form of state-dependent or demand curve. The producer reacts to time-dependent rules. The other direc- changes in demand by changing out- tion is to study the sources of sticky mar- put while leaving the relative price con- ginal cost. stant. C. S,s State-dependent Pricing Rules Only with an upward sloping MC schedule (due either to decreasing re- The new menu-cost literature owes turns to labor in production or to an in- its origins to a paper by Barro (1972) on creasing marginal disutility of work) does the S,s approach to price adjustment by the producer desire to change the rela- a profit-maximizing monopolist who faces tive price in response to a shift in de- a lump-sum cost of adjusting prices. The mand. However, because there is com- common theme linking the older S,s lit- plete symmetry across producers, erature and the newer menu-cost litera- relative prices must all be equal to unity. ture is that price setters do not change An attempt to decrease relative price in price every time the desired price level response to a decline in demand leads changes, but only when the desired level to a decrease in all nominal prices and deviates by more than a particular per- in the aggregate price level, and this ad- centage from the current price. In the justment of the aggregate price level con- S,s literature the width of the percent tinues until all relative prices are back band is arbitrarily given, while in the to unity. Money is completely neutral. menu-cost literature the width, while The only element introduced by monop- also given, is presumed to be "small" and olistic competition is a declining marginal ultimately capable of being explained by revenue schedule, which means that in particular adjustment costs. For exposi- equilibrium (with PilP = 1) price is above tory purposes these contributions may be marginal cost rather than equal to mar- discussed together, because they both ginal cost, and output is lower than in concern barriers to the adjustment of competitive equilibrium. nominal prices and share the common theme of a percentage band within which 37The model is presented in slightly simplified the price remains fixed. form in Blanchard and Fischer (1989, pp. 376-81) The basic S,s result is derived for a and Fischer (1988, pp. 321-23). An even simpler version with constant marginalcost is presented by monopolist facing a stochastic additive Rotemberg (1987, pp. 78-80). shift in its demand curve taking the form 1144 Journal of Economic Literature, Vol. XXVIII (September 1990) of a random walk without drift. The opti- trality of money. If firms face both local mal for the monopolist is shown and aggregate shocks, their price changes to be the selection of "floor" and "ceiling" will be independent and staggered across bands, with the price remaining constant time. But when they do increase their when the shift is within the bands and individual price, they will raise it suffi- changing fully to the new desired level ciently to boost the aggregate price level when the shift is outside the bands. The by the full amount of the aggregate width of the band, expressed as a per- shock. For example, if demand increases centage of the current price, depends in a series of one-unit steps, and adjust- positively on the cost of a price change ment costs limit individual firms to a and inversely on the of price increase every fourth step, then not changing, which in turn depends on that individual price increase will be four the slopes of the demand and cost units and will increase the aggregate functions. 38 price level by one unit. This result is specific to a demand dis- The Caplin-Spulber result is contin- turbance that is modeled as a random gent on an unrealistic assumption, that walk, so that changes in the disturbance the desired price follows a continuous are serially independent, and as yet opti- and monotone path. A more general mal rules have not been derived for more model, which reverses their main conclu- general processes in which the changes sion, has been developed by Caplin and in the disturbance are serially correlated John Leahy (1989). Their main point is (as surely they must be in view of serial that when the monetary shocks are two- correlation in changes in nominal de- sided, that is, when money can go both mand evident in Figures 2 and 3). In- up and down, without any monotonic stead, most of the extensions of the S,s tendency in a single direction, there can approach concern inflation which is at a be long periods in which the aggregate sufficiently rapid rate that the price level price level does not change in response cannot decrease, so the choice problem to monetary disturbances. Intuitively, is simplified to choosing the timing of money is neutral only when the economy price increases. Eytan Sheshinski and continuously hits an upper S or lower s Yoram Weiss (1977, 1983) show that the band, but a more general stochastic pro- S,s approach carries over to inflation; cess for money may leave it inside both now the price is increased at any point bands for substantial periods during when it sinks below the optimal price which there is no incentive for any agent by an amount exceeding a lower s to change its nominal price. band. A difficulty in the S,s literature is that Andrew Caplin and Daniel Spulber for analytical tractability all firms are (1987) have investigated the implications identical, and thus have price increases of aggregating S,s behavior from the level of equal size that differ only in timing. of the individual to the aggregate econ- This is belied by virtually all evidence omy. Their striking result is that one- on cross-section pricing behavior, includ- sided S,s rules (as are appropriate in an ing the differing cyclical responsiveness economy with an inflationary bias) do not of prices across industries in the Great lead to price stickiness or the non-neu- Depression (shown in Table 2). This evi- dence suggests that elements beyond simple state-dependent pricing rules 38 The original result was derived by Barro (1972) and is restated by Blanchard and Fischer (1989, pp. must lie behind observed price behavior 402-05). at the micro level. Gordon: What Is New-Keynesian Economics? 1145

D. The Menu-Cost Insight and Its to the more economically rational behav- Limitations ior of replacement cost pricing as a conse- quence of the perceived costs of delegat- Taking the S,s literature as a point ing pricing authority to lower levels of of departure, what new insights have , in contrast to general been contributed by the menu-cost liter- FIFO-type rules of thumb that save these ature developed in the mid-1980s? The costs of delegation even if they lead to menu-cost approach began defensively in pricing decisions that may be otherwise response to those critics who argued that suboptimal. All these physical costs of costs of changing nominal prices are printing, negotiating, and delegating much too small to justify output fluctua- are doubtless present in the real world tions of the size observed in the U.S. of business, although one can quibble Its key insight is that second-order ad- with their importance. Costs of negotiat- justment costs may have first-order social ing are also a key ingredient that consequences, simply because profit motivates staggered contracting, a time- functions are flat on top.39 Following a dependent rule considered in the next change in demand there may be little section. difference in the firm's profit if it does Whatever the nature of the menu or does not adjust its price, and thus even costs, the analysis may be presented in small menu costs may potentially dis- terms of Figure 3 above, which already suade the firm from price adjustment. provides the ingredients necessary to il- Yet the social consequence of such a fail- lustrate the point initially made by ure to adjust price may be large swings George Akerlof and Janet Yellen (1985) in output. and Mankiw (1985). Following Mankiw, The proponents of the menu-cost ap- we examine the situation in which de- proach are quick to admit that this widely mand has declined in Figure 3 from Do used label is misleading. Included among to D1 and marginal cost has declined from the nominal costs of price adjustment are MCo to "Required MC1." The optimal not just the literal application of the label price-output point is at D, and we ask to changing prices on menus, lists, cata- what difference is made if the firm leaves logs, and other printed material, but its price unchanged at P0. Figure 4 copies more generally the entire range of costs the new demand curve and shows the that managers must incur whenever same points B and D as in Figure 3. The nominal prices are changed. Meetings, difference between profits at points D phone calls, and trips to renegotiate with and B is shown by the rectangles T - suppliers all fall under the rubric of menu R. However, at point B total surplus is costs. Included in this more general defi- smaller by the area S + T than at point nition of menu costs is Okun's (1981) D. But the firm will only reduce price analysis of the product market. Okun ex- if the extra profit T - R exceeds the plains the reluctance of firms to shift from menu cost. Mankiw shows that as the FIFO (first in first out) pricing policies price elasticity of demand varies from ten to two, the ratio of the to the 3 Laurence Ball disputes this interpretation and profit increment varies from 23 to 200. claims that "the central point [of recent work] is that nominal rigidity has negative externalities because His results, as do Figures 3 and 4, as- it exacerbates fluctuations in real aggregate demand." sume that the marginal cost schedule is But this is clear as a matter of definition (see equation flat. In general, the flatter is the marginal (4) in Section II.A), is common to any theory of price stickiness, and has nothing to do with the particular cost schedule, the smaller are the menu contributions of recent work. costs needed to make the firm's fixed- 1146 Journal of Economic Literature, Vol. XXVIII (September 1990)

p values.40 This argument of Laurence Ball and David Romer (1989a) greatly weak- ens the appeal of the menu-cost ap- Po proach, although their own model im- plies that the second-order social costs R: ------can be much larger than the costs of P2 -- changing price. 2. But independent of the Ball-Romer T symmetry argument, the menu-cost ap- MCI proach seems flawed from the start, be- cause it considers only costs of price ad- justment and totally ignores costs of output adjustment. This places its as- sumptions in diametric opposition to Qo Q other important branches of macroeco- Figure 4. nomics, such as Tobin's q theory of in- vestment based on time-dependent price decision optimal and hence to cre- physical costs of changing the capital ate an output response from a change , or the production-smoothing the- in nominal aggregate demand. ory of behavior based on the At least four important criticisms of the assumption that a smooth rather than menu-cost approach may be offered. variable production minimizes cost. Taken together, they make a strong case Costs of output adjustment raise the cost against this core contribution of the new- of not changing price and tilt the firm's Keynesian macroeconomics. decision toward price flexibility; whether 1. A consideration of symmetry brings costs of output adjustment raise the social the basic conclusion into question. If the costs of aggregate output fluctuations de- failure to reduce price in response to a pends on the relative size of the private demand reduction makes output too low, and social costs. then the failure to raise price in response 3. The two-period comparison of Fig- to a demand increase makes output too ure 4 neglects the calculus of costs and high. Yet, starting from an initial profit- benefits in future periods. The proper maximizing equilibrium level of output setting is dynamic, as with the analogous like Qo under monopolistic competition, question of the "sacrifice ratio" in the society gains from additional output be- form of a temporary aggregate output loss cause price is above marginal cost. Hence required to achieve a permanent reduc- the menu-cost model fails to prove its tion of the aggregate inflation rate. The point: Social costs in recessions are bal- proper comparison is between the one- anced by social gains in booms. Any cost shot menu cost and the present value of from price rigidity must involve increas- the infinite stream of losses by maintain- ing the variance of output, not changing ing the price (and output) levels at differ- its mean, and hence are likely to be sec- ent values than the social optimum. The ond-order, just as the costs of changing Ball-Romer symmetry argument vitiates price are second-order. One cannot con- the force of this criticism, because the clude that sticky prices necessarily re- duce welfare, for the comparison of two 40 This important point credited to Ball and Romer, second-order effects turns on model-de- whose paper was written in 1986, is summarized and pendent comparisons of parameter endorsed by Rotemberg (1987, pp. 83-85). Gordon: What Is New-Keynesian Economics? 1147 infinite stream of losses when the price Roberts, David Stockton, and Charles is set too high is balanced by a similar Struckmeyer (1989) have found in time- stream of gains when the price is set too series data for 20 two-digit U.S. indus- low. tries over 1958-83 that the adjustment 4. Like the S,s approach, the menu- of nominal prices to nominal labor and cost approach fails to explain why prices materials costs takes place extremely rap- of some products are more flexible than idly. For four industries, 90 percent of others over the business cycle. The fail- price adjustment occurs within the ing of the Mankiw version illustrated in month, and for no industry is the first- Figure 4 is in this regard similar to that month adjustment lower than 45 per- of the canonical Blanchard-Kiyotaki cent. This provides strong evidence that model described earlier; marginal cost is the menu-cost approach is on the wrong simply assumed to move in proportion track, and that the key issues concern with demand. Once we consider the the stickiness of both wages and materials many layers of heterogeneity of products costs, not final goods prices. The mere and industries studied in the industrial fact of imperfect competition and price organization literature of Part III.A, we tags appears to be quite compatible with recognize that no individual firm can as- nominal flexibility. sume its marginal cost will be perfectly E. Time-dependent Rules and Staggered correlated with aggregate demand. Sub- Contracts sequently this will lead us to the input- output table as an essential component The last element to be considered in in the description of price stickiness and new-Keynesian explanations of nominal will reinforce our previous point that the price rigidity is the staggered contract failure to consider heterogeneity and ag- model. As noted above, new-Keynesian gregation issues is a central flaw in repre- economics can be said to begin with the sentative agent modeling. Fischer (1977a) and Phelps-Taylor (1977) Thus we return to the original objec- models of staggered contracts, which em- tion to models of nominal rigidity based phasized wage contracts. More recently, on adjustment costs. Any satisfactory models of staggered price contracts have model of price rigidity must be able to been developed by Blanchard (1983, 1986) cope with the Great Depression, yet the and Ball and Romer (1989b), among others. magnitude of demand shifts between These models investigate the implications 1929 and 1933 would seem to swamp any of staggered overlapping price-setting in- reasonable guess as to the magnitude of tervals of constant length, and in the case S,s bands or menu costs. And one does of Ball-Romer investigate the conditions not have to dip into history to doubt the necessary for firms to engage in staggering. relevance of such adjustment costs. Ev- The staggering and overlapping of inter- eryone has witnessed the fast-changing vals in which individual prices or wages price tags in the produce section of the are fixed introduces a critical element of neighborhood supermarket: There seems realism into new-Keynesian economics. As to be nothing to prevent the price of a shown by Blanchard (1983), a change in pint of strawberries from moving from nominal demand can affect output for a pe- $1.89 to $0.59 to $1.89 in successive riod that exceeds the length of the interval weeks. Carlton's evidence shows not just during which prices are predetermined, that prices can jump by large amounts which we will call the contract interval in successive weeks, but by small even though there is no necessity that ex- amounts in successive months. John plicit or implicit contracts be involved. 1148 Journal of Economic Literature, Vol. XXVIII (September 1990)

Consider contracts of n months, with a frac- rather than complete synchronization. Ball tion 1/n of agents resetting contracts each and Romer (1989b) show that staggering month. The firms that reset their price in is a stable equilibrium if there are firm- the first month following a demand shift specific shocks that arrive at different times move their price not to the optimum given for different firms. However, they show the level of nominal demand, but to the that synchronization can also be an equilib- optimum contingent on the fact that the rium: "Multiple equilibria are possible be- other outstanding contracts cause a fraction cause there is an incentive for synchronized (n - 1)/n of the aggregate price level to be price setters to remain bunched, but not preset. Any firm adjusting all the way for staggered price setters to move toward would cause a suboptimal divergence of its synchronization" (1989b, p. 193). There relative price from the optimum level, seems to be a debate as to whether firm- given the stickiness of other prices. specific shocks are sufficient to guarantee The study of staggered prices takes as a staggering equilibrium, but only in the its point of departure that the length of context of simple models in which firms predetermination of prices reflects a bal- can choose only to change price at odd or ancing of the costs of adjusting prices and even dates. A more general setting in the opportunity costs of nonadjustment, which some prices are changed weekly and just as in the S,s model. Because this deci- others yearly, and in which the yearly price sion need be made only by the profit-maxi- changers have 365 possible dates on which mizing price-setting monopolist, there is to change, destroys the argument that each no need for an actual contract with another individual price setter still has an incentive agent. There has been little attention to to "bunch" price changes at the same time, the nature of the adjustment costs or in because there is no such thing as the "same particular their variation across industries time. 41 or over time, which is unfortunate as this might provide the element that is missing VI. Sources of Real Rigidity in the Product in so many new-Keynesian models, the Market ability to explain cross-time and cross-in- A. Customer Markets dustry differences in price behavior. In particular, there is no element in the the- The analysis of nominal price rigidity ory that would explain why the rate-of- in Part V treats only the first quadrant of change (ox)coefficient of price adjustment our two-by-two matrix defined along the increased in most countries during World 41 For this reason I find unconvincing the skepti- War I, or why the inertia (A) effect in- cism of Blanchard and Fischer (1989, p. 401) that it creased in most countries except Japan af- is possible to derive stable staggered contracting, as when they write "the introduction of stochastic idio- ter World War II. As a separate criticism, syncratic shocks does not make staggering more there has been no attempt to introduce ex- likely." Their argument is carried out entirely within plicit indexation into these staggered con- an either-or choice between even and odd dates of price changing, and they show that a 50-50 equilib- tract models, which contain no element to rium with the same number of firms choosing each explain why firms do not predetermine real date is unstable, because the slightest tilt in either prices for a time interval (presumably to direction gives all the other firms an incentive to shift. But with uneven frequency of shocks and a save on management decision costs) and large number of possible dates of changing, the incen- then index the nominal price to nominal tive to shift disappears. If my optimal frequency of GNP. price change is weekly, the fact that there is bunching with more price changes on January 1 than any of Instead, attention has been concentrated the other 364 days of the year does not lead me to on the question of why there is staggering limit my price changes to once a year on January 1. Gordon: What Is New-Keynesian Economics? 1149

dimensions of product versus labor market premium to do business with customary and nominal versus real rigidity. We turn suppliers, and intertemporal comparison now to models of real rigidity, that is, mod- shopping discourages firms from changing els that explain why real wages or prices prices in response to short-run changes in are unresponsive to changes in economic demand in order to avoid giving customers activity. In the product market, a model an incentive to begin exploring. Okun ar- of real rigidity explains why a firm would gues that his customer-search model ex- choose to hold its relative price or price- plains markup pricing practices based on cost margin constant. In the context of the full costs. Customers appear willing to ac- canonical Blanchard-Kiyotaki cept as fair an increase in price based on model, this occurs with a constant margi- a permanent increase in cost, whereas tran- nal disutility of work and constant re- sitory events, whether an increase in de- turns to labor. In the textbook monopoly mand or a reduction in productivity, are model of Figure 3, the price-cost margin not generally expected to last long enough is fixed if the marginal cost schedule is to justify price increases. horizontal and the elasticity of demand is Okun's approach has several critical de- constant. fects. He argues that price increases based Recall that our discussion of indexation on cost are perceived as fair, while cost in Part IV. C introduced a basic objection increases based on demand are not so per- to all models of real rigidity. No matter ceived. The case for customer dissatisfac- how rigid is the real wage or real price, tion is difficult to argue, because any loss what prevents the nominal price and wage of goodwill created by a price increase in from being indexed to nominal GNP? At a boom would be balanced by a gain of that point we asked whether nominal rigid- goodwill created by a price decrease in a ities were sufficiently important to be able recession. Okun seems to be thinking of to explain the absence of nominal GNP in- an inflationary world in which price dexation. Thus far we have concluded that changes are one-sided, as in the trend-in- nominal rigidities based on S,s or menu- flation S,s literature. Further, the fairness cost models are not convincing, while time- explanation leaves open the determination dependent rules in the form of staggered of fair behavior, and in fact what is per- price-setting intervals are completely com- ceived to be fair may just reflect whatever patible with any form of indexation. Thus behavior may be normal, for whatever rea- a critical test for the theories of real price son. Thus Okun's approach has an element and wage rigidity is whether they stand of circularity. up to the indexation criticism. Okun's approach also seems vulnerable Perhaps the earliest prominent model of to the same criticism often directed at Lu- real rigidity in product markets explains cas-type new-classical models: Why should why customers do not respond instanta- a firm be afraid to lose customers when neously to changes in real prices, that is, raising prices in response to higher nominal in the price charged by one firm relative demand if information on higher nominal to others. Arthur Okun (1975), building on demand is instantly available? What pre- the work of Armen Alchian (1969) as well vents all firms from indexing to nominal as Phelps and Sidney Winter (1970), popu- demand and price specials on larized the distinction between auction and items priced lower than would be war- customer markets. The former are per- ranted by the indexation formula? As we fectly competitive. But, in the latter, costly have seen already, the elementary theory search makes customers willing to pay a of monopoly pricing behavior by itself sug- 1150 Journal of Economic Literature, Vol. XXVIII (September 1990) gests little for price flexibility. Everything shocks, and local and aggregate cost depends on the response of marginal cost shocks.42 to aggregate demand shocks. This four-way distinction creates two complementary sets of reasons why firms B. The Independence of Costs and may rationally expect marginal cost to Demand move differently from marginal revenue. Nevertheless, there is a deeper insight First, marginalrevenue may move with ag- in Okun's distinction between cost and de- gregate nominal demand but marginal mand. Firms raise price in response to an costs may not. This would occur if a firm upward shift in the marginal cost schedule believes that its costs depend not just on not just because it is optimal in a textbook nominaldemand but on local supply factors model, but because they will go bankrupt (e.g., harvests, strikes, price changes for if cost rises sufficiently in relation to price. imported materials). Second, in a situation There is no such economic necessity of rais- with nominal aggregate demand fixed, a ing price in response to an increase in de- firm might face a local shift in demand mand when cost is fixed, and for a mo- (e. g., a decline in beer drinkingin response nopolist such a price increase is not even to drunk-driving ) that reduces mar- optimal with a constant-elasticity demand ginal revenue, while marginalcost is fixed, curve and a flat MC schedule. When tied to aggregate nominal demand. More OPEC raises the price of oil sharply in rela- generally, any set of covariancesamong the tion to the level of nominal aggregate de- four shock concepts is possible. mand, everybody understands why the lo- C. The Role of the Input-Output Table cal service station raises the price of gasoline at the pump, but they do not un- To be a credible explanation of real derstand why an increase in aggregate de- price rigidity, the distinction among local mand requires any such response of the and aggregate cost and demand shocks gasoline price if the costs of service station must be embedded in a world with many inputs are fixed. heterogeneous firms interacting within a This distinction hinges on the possibility complex input-outputtable. With only two that shifts in marginal cost can be indepen- firms, each supplying the other, informa- dent of shifts in aggregate demand. Our tion would be cheap enough to permit both historical study of price adjustment in Parts firms to disentangle the local versus aggre- II and III stressed the theoretical and em- gate component of their costs. But with pirical importance of supply shifts. Ball and thousands of firms buying thousands of Romer's (1989b) study of staggering em- components, containing ingredients from phasizes the critical role of idiosyncratic many other firms, the typical firm has no firm-specific shocks. Giuseppe Bertola and idea of the identity of its full set of suppliers Ricardo Caballero (1990) emphasize the when all the indirect links within the input- role of idiosyncratic in explain- output table are considered. Because the ing infrequent price adjustment at the mi- informationalproblem of trying to antici- cro level. New-classical Mark I macroeco- pate the effect of a currently perceived nomics was built on Lucas' distinction nominal demand change on the weighted- between local and aggregate shocks. averagecost of all these suppliers is difficult Okun's emphasis on cost-based pricing to formulate and probably impossible to leads us to broaden Lucas' two-way distinc- tion between local and aggregate demand 42 I have previously (1981, pp. 520ff.) suggested a distinction between aggregate and local components shocks and suggest a four-way distinction of both cost and demand with explicit reference to between local and aggregate demand Lucas' original two-way classification. Gordon: What Is New-Keynesian Economics? 1151

solve, since (as Bresnahan emphasizes) mestic nominal demand would occur to any thousands of elasticities are involved, the from, say, or . Be- sensible firm just waits by the mailbox for cause we know that par- news of cost increases and then, Okun-like, ity (PPP) fails and that real exchange rates passes them on as price increases. are volatile, why would any firm adopt in- The input-output table approach pro- dexation of its price to domestic Belgian vides a critical contribution not just to un- or Chilean nominal GDP, which would dis- derstanding real price rigidity, but also connect its price from the large share of nominal rigidity. The standard accusation its costs that are imported? The input-out- against all theories of real rigidity, made put approach, by stressing the indepen- often above, is that they are consistent with dence of marginal cost and aggregate de- nominal flexibility achieved through index- mand, provides an understanding of the ation to nominal demand. Yet the input- lack of indexation to domestic nominal output table approach emphasizes the high GNP and thus the critical link that converts fraction of a firm's costs that is attributable a theory of real rigidity into a theory of to suppliers of unknown identity, with nominal rigidity.43 some unknown fraction producing in for- A firm's viability depends on the relation eign countries under differing aggregate of price to cost, not price to nominal GNP. demand conditions. This environment Aggregate macroeconomic stability is a would give pause to any firm considering subject to a free-rider prob- nominal demand indexation of the product lem. No individual firm has an incentive price, because the failure of all suppliers to take the risk posed by nominal GNP to adopt similar indexation could lead to indexation, which would take away from bankruptcy when nominal demand de- the firm the essential control required of clines. Thus the input-output approach the relation of price to cost. In this sense, borrows one element that is basic to the input-output explanation of nominal Keynesian economics, the coordination rigidity requires capital markets that are failure that arises from the lack of private imperfect enough to penalize the profit vol- incentives to solve a social problem, with atility that would result if a firm tried to another element inherited from Lucas, the index its prices to nominal demand without distinction between aggregate and idiosyn- being sure in advance that its suppliers cratic shocks. would do likewise. One criticism of the input-output ap- There is another sense in which the in- proach claims that with perfect information put-output table explains nominal rigidity. about aggregate variables, the only equilib- It creates a technological environment for rium of the economy would be for immedi- staggered price setting, similar to but more ate adjustment of all prices to nominal complex than Taylor's staggered wage set- shocks. Yet this ignores the fundamental ting. Today's product price is based on assumption that marginal cost and marginal costs set at many different dates in the past revenue are imperfectly correlated with ag- as product components weave their way gregate demand. Under these conditions through the input-output table. This may each firm would be unwilling to index price appear to violate the maxim that prices to nominal GNP both because marginal should be based on replacement cost. But cost may not move with nominal GNP even there are too many links in the input-out- if marginal revenue were to do so, and vice versa. 43 Cooper (1989) provides an analysis of the inter- dependence between wage and price indexation; the A good reason for every domestic firm likelihood of wage indexation depends on whether to refuse to index its product price to do- prices are indexed, and vice versa. 1152 Journal of Economic Literature, Vol. XXVIII (September 1990) put table for the producer even to guess Thus the input-output approach is comple- what the replacement cost may be. The mentary to theories of rigidity in the labor automobile firm may receive a notice from market. the headlight maker of a price increase, A compromise place to end the dis- but no warning of a price-increase notice cussion of product-market rigidities is to that is alreadyin the mail from the filament admit that the input-output approach is maker to the headlight maker or from the complementary as well to at least one of maker to the filament maker. the new-Keynesian approaches based on Blanchard (1987b) uses the term cumula- nominal rigidities. The input-output ap- tion hypotheses to describe the role of the proach needs some additional element to input-output table in translating prompt explain why we do not observe in the real price adjustment at the individual level to world extremely frequent small price gradual price adjustment at the aggregate changes every day as firms react to each level. He provides suggestive supporting tiny cost change as it arrives in the mail evidence that in disaggregated data prices through the input-output table (while Carl- adjust faster than in . The ton documents some such small changes, automobile, headlight, filament, and cop- long intervals of complete rigidity are com- per makermay all respond to cost increases mon as well). within a day, but months can separate the The core element that needs to be added effects of a change in the price of copper to the input-output approach is a cost to from the ultimate change in the price of making price changes every day that causes automobiles. rational managers to concentrate price- The input-output table approach domi- setting decisions at discrete intervals. nates menu costs in explaining why the There is no need to force this sort of nomi- price of strawberriesis more volatile than nal rigidity into a single semantic category; the price of automobiles(because strawber- the core factor for some firms may best ries are not physically transformed from be described as staggered time-dependent farm to market). It explains the different rules and for others as state-dependent rate-of-change adjustment coefficient (x) rules based on menu-type costs. Undoubt- across industries by two auxiliaryassump- edly these categories overlap because tions. First, auction markets are distinct many firms face both time-dependent and for customer markets and are limited state-dependent costs. Some firms that mainly to crude and intermediate goods. routinely hold price-setting meetings once Thus products like strawberries and plas- a week or month to save on managerial tics that appear relatively early in the in- time costs may decide at those periodic put-output chain have relatively flexible meetings to leave some or all prices un- auction-likeprices. But what is it that cre- changed when the difference between the ates more price rigidity for more complex current and optimal price does not yet ex- products later in the chain? Partly it is the ceed the perceived cost of printing new of large numbers that cancels out idio- menus and catalogs. syncratic supply shocks for final products incorporating large numbers of different VII. The Search for Structure: Labor purchased materials. But there also may Market Behavior be a role for wage rigidity, as the prices A. The Relation of Wage and Price of products embodying relatively large amounts of embodied labor, like automo- Behavior biles, tend to be more rigid than that of The dominant new-Keynesian view is products embodying large amounts of em- that nominal rigidities originate in the bodied , like wheat or strawberries. product market, not the labor market. The Gordon: What Is New-Keynesian Economics? 1153

path of wages, to use the words of Mankiw, have the very un-Keynesian implication is "completely indeterminant and com- that in recessions workers are close to indif- pletely irrelevant" (1988, p. 446). Yet ferent between working and not working." surely this goes too far. Mankiw follows For new-Keynesian models to avoid in- earlier writing, notably by Barro (1977b) consistency, their distinction between and (1980), who have argued small menu costs of price changes and large that wage rigidity is irrelevant for employ- social costs of output changes must apply ment determination. In the context of a equally in the labor and product markets. long-term or even lifetime , there is no The same costs of adjustment that inhibit reason for the wage in a given time period price changes must apply equally to wages, to be equal to the marginal product of la- which are just another price. Sticky prices bor. The wage can be an installment pay- cause changes in nominal aggregate de- ment on a lifetime contract. mand to be transmitted directly to shifts However, the claim that sticky wages are in the demand curves facing not just indi- irrelevant to allocation calls for prices to vidual firms, but also individual workers. be perfectly flexible, as is required for per- The Barro-Grossman model dis- fect market clearing, while wages are cussed in Part IV. B achieves the desired sticky. We have already observed in Sec- symmetric treatment, in which sticky tion II. C that capital markets are likely to wages and prices cause both firms and impose a tax on the resulting profit variabil- workers to face constraints on the amount ity. Further, the monopolist example they can sell. shows that prices will not be perfectly flexi- Blanchard and Fischer (1989, p. 427) ble unless all elements of marginal cost are state that the key issue in new-Keynesian perfectly flexible. This brings us back to economics is explaining why "labor and indexation: The sticky wages that are in- output supply functions [are] relatively stallment for lifetime must flat." They intend this phrase to mean real be fully indexed to nominal demand for rigidities.44 Yet their choice of words is un- Barro, Hall, and Mankiw to be correct that fortunate, because it ignores the distinction sticky wages are irrelevant to allocation. between aggregate and individual supply Just as it is implausible for wages to be curves, as well as between notional and sticky while prices are perfectly flexible, effective supply functions. The labor sup- so is the reverse, for wages to be perfectly ply function of an individual head of house- flexible while prices are sticky. Yet this is hold may be vertical, but any mechanism just what is assumed in much of the menu- that rigidifies the real wage will cause the cost literature reviewed in Part V. When individual to be pushed off of this notional menu costs lead rational firms to avoid supply function. Actual behavior traces price changes and meet demand through shifts in the effective labor demand sched- changes in output, corresponding fluctua- ule and tells us nothing about the shapes tions in labor input are required. In menu- of notional functions. In our interpretation cost models the real wage adjusts to make the key issue is the explanation of wage workers willing to change the amount they and price rigidity, not the explanation of work; that is, the nominal wage rate is per- fectly flexible. In the Blanchard-Kiyotaki canonical model of monopolistic competi- 4 Blanchard has written to me that what he means by flat labor supply is "the set of real wages and tion, the representative agents set their rel- employment traced out as the marginal product of ative price to minimize the marginal dis- labor shifts" and not "the competitive labor supply utility of work; that is, they slide along their curve." The issue here is the possibly misleading choice of words, not any substantive difference be- voluntary labor supply curve. As Rotem- tween my interpretation and that of Blanchard and berg has noted, "both of these approaches Fischer. 1154 Journal of Economic Literature, Vol. XXVIII (September 1990) why labor and output supply curves are croeconomics volume was not to business flat.45 And at a deeper level, as argued cycle theory but rather to explain why the above in Section IV.B, the really central naturalunemployment rate is greater than element is the coordinationfailure that un- zero, due chiefly to the work of Dale Mor- derlies wage and price rigidity. tensen (1970a, 1970b). In a world of costly information and heterogeneous jobs and B. Early Theories of Real Rigidity: Search workers, workers sample from an Models and Implicit Contracts array of job offers and firms sample from an array Just as theories of price stickiness can of workers. Unemployment is a voluntary usefully be divided between theories of activity, but all voluntary unemployment nominalversus real rigidity, so can theories is not socially beneficial, and government of wage stickiness. Reflecting the chrono- tend to stretch out logical development of the field, we begin the interval between searches, imposing a with models of real wage rigidity and then social cost throughthe levied on some turn to models of nominal wage rigidity. to support the extended search interval of Widely recognized as the first attempt others. The new microeconomics volume to build structuralmodels of labor market also contained the important Phelps-Win- behavior as the outcome of maximizingbe- ter (1970) theory of customer markets, havior was the new microeconomicsof the based also on the assumption of imperfect famous volume edited by Phelps (1970). information. We have seen that this was Most of the papers in the Phelps volume, later picked up and developed by Okun including Phelps' own desert island para- in a new-Keynesianrather than new-classi- ble, yielded market-clearing conclusions cal setting. and as such should be regarded as part of While the new microeconomicswas ex- the development of new-classical rather plicitly classicalin approach,the next wave than new-Keynesian ideas. In the parable, of contributions under the heading of im- workers are on isolated islands and react plicit contract theory was the first to de- to a wage cut by boarding rafts to sample velop what some initially thought was a wage offers on other islands. Variationsin microeconomic explanation for Keynesian employment during business cycles are wage stickiness. In the simultaneously due solely to the voluntary response of written and independent contributionsby workers to changes in the expected real (1975), Martin Baily wage. The parable ignores the prompt (1974), and Donald F. Gordon (1974), em- availability of aggregate information, fails ployees were assumed to be relatively to explain layoffs and "no help wanted" more risk averse than their employers, signs, and yields the counterfactualimpli- mainly because of self-selection of individ- cation that voluntary quits vary counter- uals to become entrepreneurs. Firms maxi- cyclically (see Okun 1981, ch. 2). mized profits by minimizing the variability The main contribution of the new mi- of income to their workers, who disliked variability,in effect providing a compensa- 45 Thus I concur with Barsky and Solon (1989, pp. tion package that consisted partly of pecu- 29-30), who find that procyclical real wage behavior niary wage payments and partly of insur- at the individual level in micro data is consistent ance services. with noncyclical behavior in aggregate data. This pat- tern reflects a cyclical variation in the "employment It was soon recognized that this approach opportunities" (read "constraints") that face both provides no satisfactory explanation of "stayers" at firms who face changing opportunities Keynesian unemployment; it justifies only to work , and "switchers" who face cyclical- ity in opportunities for across-firm advance- a fixed-incomecontract (i.e., tenure) rather ment. than the fixed-wage variable-employment Gordon: What Is New-Keynesian Economics? 1155

contracts actually observed. Variable em- Ian McDonald and (1981), ployment is explained only by a gratuitous who show that a be- element patched onto the theory, govern- tween a firm and a union can lead to rela- ment side-payments during periods of un- tively large employment fluctuations and employment. Even when these variable relatively small real-wage fluctuations, thus employment fixed-wage contracts are gen- contributing a source of real-wage rigidity. erated by the theory, they have the un- In an extension, McDonald and Solow Keynesian implication that workers are (1985) examine the impact of business- equally happy when employed and unem- cycle fluctuations on a labor market seg- ployed.46 Further, workers are shown mented into a union primary sector and to care about stability in real income, a competitive secondary sector. Reflect- not nominal income, so implicit contract ing the small real-wage fluctuations in the theory has no explanation for the failure union sector, they show that either per- of workers to insist on full indexation of manent or temporary changes in real ag- wage contracts. gregate demand widen sector wage dif- ferentials in recession and cause greater C. Labor Unions fluctuations in primary sector than second- The effects of labor unions have been ary sector employment. extensively analyzed by labor economists. Yet the formal theory of unions does not Bargaining models have been developed provide a general explanation of Keynesian in which firms and unions, which in turn wage rigidity. If union members care about act on behalf of their member workers, bar- stability of employment, it is difficult to gain over wages and employment. Some understand why they are willing to tolerate models characterize the employment deci- a wage rate that is set for a substantial inter- sion as a unilateral decision of manage- val, while the decision on the amount of ment, as it is in many contracts. These employment is left to the firm.47 Obviously models that are concerned only with wage if the wage rate is predetermined as part setting are sometimes called the right-to- of a union contract, this rigidifies marginal manage model and fall between two ex- cost and hence prices, and nominal de- tremes. At one extreme firms are all power- mand fluctuations are transmitted to out- ful and are able to pay the put and employment. But, overall, the possible, that is, the competitive wage. Be- union literature leaves open the question cause firms have complete control of both why the wage rather than the level of employment and the wage, this subclass employment is set by contracts, and why of models does not warrant the label bar- the wage rate is not indexed to nomi- gaining model at all; the efficiency wage nal demand so as to stabilize employ- models discussed in Section VII. E fall into ment. this class. At the other extreme is the Another problem is raised by the empiri- union-monopoly model dating back to cal evidence of Table 3 in Part III. Unions Dunlop (1944); here too there is no bar- became important in the U.S. only after gaining, because firms set employment and the mid-, yet the estimated rate-of- unions set the wage. change (x) and level (y) effects for the U. S. A more general model is developed by are the same during 1873-1914 and 1954- 87. Some factor other than unions must 46 More technically, as pointed out to me by account for the price stickiness evident in Blanchard, it is the of that is equalized between the employed and unem- ployed. The ranking of utility depends on the form 47This point is made by Blanchardand Fischer of the utility function. (1989, p. 453). 1156 Journal of Economic Literature, Vol. XXVIII (September 1990)

U.S. data for the nineteenth century. The firing costs, as well as costs that can be main contribution of unionization may have directly imposed by the insiders when they been the particular American phenomenon shirk or fail to cooperate in the presence of the three-year staggered wage contract, of underbidding outsider entrants. Never- which has doubtless contributed to the theless, the insider-outsider theory does higher inertia parameter for both wages suggest a rationale for unionization by and prices. Yet here too there are prob- showing how unions can organize and lems, because inertia in price change coordinate insiders' rent-seeking activi- seems to have increased more than for ties. wage change (Table 3), and, further, inertia The insider-outsider theory sheds light increased substantially for price change in on a variety of labor market phenomena, the postwar period in the U. K., France, such as the persistence of unemployment, and Germany (Table 4), nations in which differences in variability of employment the three-year contract is not prevalent. across industries and countries, labor mar- ket segmentation, the durationand compo- D. Insider-Outsider Theory sition of unemployment, and the interin- Another body of work that deals with dustrywage structure.The theory has been the existence and persistence of unemploy- applied to the puzzle of persistently high ment is the insider-outsider theory. The unemployment in Europe in the 1980s by insiders are experienced incumbent em- Blanchardand Lawrence Summers (1986) ployees whose jobs are protected by a vari- and Lindbeck and Snower (1988) and has ety of labor turnover costs which make it become one of several explanationsof the costly for firms to replace them. The out- hysteresis hypothesis (see Section II. D), siders, who are either unemployed or work in which the rate of unemployment de- in the casual or secondary labor market, pends on the history of actual unemploy- have no such protection. Assar Lindbeck ment rather than, as in Friedman'soriginal and Dennis Snower (1986, 1988) argue version, being "groundout" by the micro- that, owing to these turnover costs, the economic structure of the economy. The insiders gain market power, which they use insider-outsider approach explains the to their own advantage, without necessarily emergence of high unemployment in the taking fully into account the of 1980s as an indirect consequence of the the outsiders. Further, the insiders often oil shocks of the 1970s, which created a can influence the turnover costs them- temporary adverse reduction in labor de- selves by agreeing to cooperate among mand and caused the insider work force themselves but not with outsiders should to contract. When labor demand recovered the latter attempt to gain employment by the remaining insiders set wages to maxi- underbidding the insider wage. This struc- mize their own welfare, thereby discourag- ture causes unemployment for the outsid- ing employment and making the high un- ers, who cannot find jobs even though they employment persist.48The best evidence would be willing to work for less than the in support of this approach is the work of prevailing insider wage. Layardand Nickell (1987)which shows that Although the insider-outsider theory the demand pressure variableentering the contributes to our understanding of union Phillips-curve wage equation is not total behavior, it is not primarily a contribution unemployment, but rathertotal unemploy- to the union literature. A wide variety of labor turnover costs may well be significant 48 To this point the discussion of the insider- outsider model is largely based on several paragraphs even in the absence of unions, for example, of text kindly contributed by Assar Lindbeck and hiring, training, negotiation, litigation, and Dennis Snower. Gordon: What Is New-Keynesian Economics? 1157

ment minus the long-term unemployed. a higher-quality labor force, and improved However, to the extent that it explains the morale and loyalty.50 Virtually all the litera- persistence of high European unemploy- ture with implications for macroeconomics ment by high insider real wages, it is sub- dates from the 1980s. Although most sur- ject to the criticism (R. Gordon 1988) that veys trace the germ of the idea back three high unemployment was immune to the decades to early work on less-developed moderation of real and the countries that posited a linkage among disappearance of the European wage gap wages, nutrition, and health (e.g., Leiben- in the 1980s. stein 1957), the terms efficiency wages and efficiency earnings appear in Alfred Mar- E. Efficiency Wage Theory shall's Principles (1920, pp. 456-69). An- If any development in the microeco- other precursor of the idea is the negative nomics of labor markets could be called relationship between wages and quit rates the "rage of the 80s," it is efficiency wage embedded in Phelps' (1970) desert island theory, based on the hypothesis that parable and other early models in the new worker productivity depends on the level microeconomic literature. Efficiency wage of the real wage. When there is such a theory provides a rare common meeting link between the wage rate and worker effi- ground for and radical econo- ciency, firms may rationally pay a real wage mists, because the far left in U.S. econom- rate that exceeds the market-clearing level. ics has taken the lead in developing theo- Firms may refuse to reduce the wage to ries of dual labor markets and for setting hire members of a pool of unemployed out policy proposals for higher minimum workers who may be available at a lower wages based on the assumed validity of the wage, fearful that a reduction in real wages efficiency wage approach.5' for existing workers may reduce productiv- The basic efficiency wage result is ob- ity by more than the gain in lower wages. tained in a simple model with identical, The appearance of an of labor perfectly competitive firms and a produc- in such a setting can be shown to be consis- tion function in which labor input is multi- tent with maximizing behavior of both plied by an efficiency factor e that depends firms and workers. There is substantial on the real wage. Because the elasticity overlap between the insider-outsider and of e with respect to the real wage declines efficiency wage models, as they both focus as the real wage increases, the first-order on barriers to underbidding by unem- conditions require the firm to choose an ployed outsiders. While the insider-out- optimal real wage rate (w*) at which this sider approach emphasizes the market power of incumbent workers, the efficiency 50 Two surveys of the literature that identify those wage approach stresses the choice prob- authors and papers who have studied particular chan- lem of firms that have imperfect informa- nels of efficiency wage effects are Katz (1986) and Weiss (1990). tion about the productivity of their em- 51 On dual labor markets, see especially Peter ployees.49 Doeringer and Michael Piore (1971) and David Gor- The reasons for the response of pro- don, Richard Edwards, and (1982). More recent evidence by mainstream economists is ductivity to the real wage vary across mod- provided by William T. Dickens and Kevin Lang els and include effort, reduced shirking, (1985). For policy proposals based on efficiency wage lower turnover and training costs, the abil- assumptions, see Samuel Bowles, , and Thomas Weisskopf (1983). Their policy proposal to ity of high-wage firms to screen and obtain raise the minimum wage assumes implicitly that the current wage is below the optimum efficiency wage, whereas all the work in the new-Keynesian tradition 4 An excellent comparison of the two approaches examines the implications of assuming that the actual is provided by Lindbeck and Snower (1988, ch. 3). wage is already at the optimum efficiency wage level. 1158 Journal of Economic Literature, Vol. XX VIII (September 1990) elasticity is unity. Workers are hired up that the same monitoring problems that to the point where their marginal product generate the efficiency wage result also equals the optimal wage (w*). The intuition make it unlikely that or other financ- of the unit elasticity result is that firms ing sources will come forth to provide fi- forgo efficiency gains that yield more than nance for the initial lump-sum payments they cost when they pay below w*, while or performance bonds. This defense does a wage above w* would cost more than it not rule out low-wage , yields in efficiency gains. Stated another which are in fact observed, but these can way, effective labor cost is minimized at be interpreted alternatively as a means of w*. sharing the cost of training rather than as Because w* is completely fixed by what- the "sale" of a job by a firm. A second ever factors of taste and technology that criticism is that the efficiency wage model determine the e function, the firm's reac- is dominated by direct payments to work- tion to any change in its relative price (i.e., ers in proportion to their efficiency, a demand shock) is to cut employment whether through piece-rate contracts or while maintaining the wage rate at w*. through 'tournaments" that pay workers Firms have no incentive to cut the actual accordingto their rankingby performance. wage, because this would actually increase The defense against this criticism is similar their wage bill per unit of output. The ex- to the first. Piece rates and tournaments treme result of a fixed real wage in this are subject to informationproblems. Work- model stems from the assumption that a ers involved in joint production do not of- worker's efficiency depends on the absolute ten have a unique claim to a "piece," while level of the real wage rather than on the payments to a team invite shirkingby some real wage relative to something else, members of the team. Tournaments are whether some measure of economy-wide also difficultto implement; there are rarely real earnings or real wages in a perceived many workers in a firm doing exactly the peer group or comparison group. A variant same tasks and no way to rankacross tasks. of this approach, in which effort depends Overall, the efficiency wage approach on the relative real wage and on the unem- seems to be an essential ingredient in ex- ployment rate (a high value of which raises plaining numerous aspects of microeco- effort by increasing the cost of job loss), nomic labor market behavior, including allows the real wage to regain some flexibil- segmented labor markets, persistent wage ity and to depend inversely on the unem- differentialsfor similarworkers that are not ployment rate (Carl Shapiro and Stiglitz equalizing differences, queues for high- 1984; Summers 1988). paid jobs, and procyclical fluctuations of Several criticisms of the efficiency wage the quit rate.53Variations on the model can approach have been offered.52 One line is explainwhy firms sometimes dismiss work- to propose that job applicants should "buy" ers instead of cutting their wage. However, high-wage jobs either by offering lump- as a theoretical underpinning of the new- sum payments or performance bonds to Keynesian paradigm, it suffers from the employers, or by offering to work in low- same defect as all models of real rigidities. wage apprentice status for an initial period. If workersgear their effortto the real wage, Efficiency wage proponents point out, however, that unemployed workers lack 53 The most controversial item on this list is persis- sufficient and are risk averse, and tent wage differentials, as argued by Katz and Sum- mers in a series of papers, including Katz (1986) and Katz and Summers (1989). For a sample of a dissent- 52 This paragraph summarizes Weiss (1990, pp. 6- ing view, see Robert Topel's comment which appears 10). after the latter paper. Gordon: What Is New-Keynesian Economics? 1159

there appears to be no barrier to full wage preceding criticism-that the negotiation indexation that allows firms simultaneously costs that rationalize the existence of con- to maintain worker effort through mainte- tracts do not rule out fully indexed con- nance of the optimal real wage w*, while tracts. The costly negotiations set the real changing the nominal wage in tandem with wage, while the nominal wage is costlessly the nominal price in order to achieve mac- indexed, preferably to nominal GNP. roeconomic self-correction. Further, the The Fischer-Taylor contract literature is efficiency wage theory has little to say set up entirely in nominal terms and does about the sources of variations in wage and not discuss the of full nominal de- price responsiveness over time and across mand indexation, so we will discuss it on countries that were identified in Part III. those terms, as a source of nominal rigidity. This negative verdict applies only if a In Fischer's version the wage for half the new-Keynesian explanation of nominal workers is set for two periods at the begin- wage and price rigidity is erected on the ning of period t and for the other half at sole base of the efficiency wage theory. t + 1. The wage set for the first group However, once the input-output approach can respond to any change in the money and the independence of local and aggre- supply in the first period but not in the gate costs and demand are accepted as the second. The greater flexibility of nominal underlying reason why actual money than of nominal wages is an assump- do not index to nominal demand, the way tion rather than a result and leads to real is open to accept the efficiency wage ap- effects of perceived monetary disturbances proach as another source of cost rigidity that cannot occur within the new-classical within the input-output table, of poten- framework. Fischer's version assumes no tially equal importance with the uncertain barriers to price flexibility and market evolution of the prices of purchased materi- clearing in product markets. The unem- als. Once again, we find that the new- ployment his model generates during a pe- Keynesian approach is most convincing riod when money has declined but wages when sources of real and nominal rigidity have not declined is classical (because of are combined rather than when either one an excessive real wage), not Keynesian. or the other is proposed as the sole In the setting of an n-period rather than explanation.54 two-period contract model, Taylor (1980) makes the nominal wage fixed over the life F. Nominal Rigidities:Wage Contract of the contract (at a level that depends on Models the expected price and expected output) and setting the price as a simple markup Theories of wage stickiness can be over the average wage rate. A monetary based on real rigidities, as in the ap- disturbance falls fully on output during the proaches outlined above, or on nominal period until the next contract renegotia- rigidities. The most influential work that tion. Then wages can adjust quite rapidly, rationalizes nominal rigidities in new- because the dependence of the negotiated Keynesian labor is the stag- wage on expected future output creates a gered-contract approach of Fischer (1977a) strong feedback loop between unemploy- and Taylor (1980). The classification of con- ment and wage behavior. Nevertheless, tract rigidities as nominal is subject to the because prices depend on wages set in any previous contract still in force, the duration of the real output response to a nominal 54In this important conclusion we fully endorse the basic message of Akerlof and Yellen (1985) and monetary shock can last for much longer Ball and Romer (1987). than the length of the contract, the same 1160 Journal of Economic Literature, Vol. XXVIII (September 1990)

result as was subsequently derived by real demand. For wage setters to use a Blanchard (1983) for the product market model to calculate the implications of their (see Section V.E). current wage-contract decisions on future Taylor'sapproach is sufficientlyplausible real demand requires not only a universal and important to take seriously. Yet it is belief that the announced disinflationary subject to at least two criticisms. First, the path of nominal demand will be maintained assumptionsof staggeringand of fixed con- on target, but also a universal ability to tract length are arbitrary. In some places forecast the response of actual prices to the (especially Japan)contract expirationdates path of nominal demand, as is required for among firms in the union sector are nearly future real demand to be predicted. simultaneous. If contract length depends Where then do staggered wage contracts on a balancing of negotiation costs and the fit in? We have seen that full price flexibil- allocative costs of infrequent adjustment, ity for a monopolist requires full flexibility one would expect changes in contract of marginal cost, and staggered contracts length in response to the variability of ei- eliminate that full flexibility in the absence ther local or aggregate shocks. A more gen- of instantaneous nominal GNP indexation. eral statement of this first criticism is that Barro, Hall, and Mankiw have argued that the existence of nominal wage contracts is it is possible for firms to adjust their prices not explained from the first principles of in proportion to a change in nominal aggre- microeconomics. Models of optimal con- gate demand if wages do not adjust. But tracting do not produce the nominal sticki- this is not profit maximizing. In almost any ness generated by the Taylor-type con- model of monopolistic price setting, an in- tracting models. The Taylor approach complete adjustment of wages implies less needs to be supplemented by an extension than full adjustment of profit-maximizing to wage setting of recent work on staggered prices. In this sense, recent new-Keynes- price setting by Ball and Romer (1989b) ian theorists have gone overboard in shift- and others, as reviewed above. ing the emphasis from the labor to the The second problem is that, once the product market. contract expires, the adjustment of the wage in response to expected future output VIII. Conclusion is not complete but is bounded in Taylor's model by an arbitraryPhillips-type adjust- We have stressed throughout the need ment coefficient. As argued by Blanchard for new-Keynesian theory to address the (1987a),Taylor's results require this adjust- most important elements of variability in ment coefficient to be relatively "small" the adjustment of prices along three di- and, if a cyclical response of the markup mensions, the inertia effect (A), the rate- of prices over wages is allowed, that must of-change effect (ox), and the level effect be "small"as well. While Blanchard'spoint (-y). The literature suggests that Taylor's wage adjustment contributes ample evidence of differences may be too slow, I would argue the oppo- in rate-of-change effects across industries. site. In particular, Taylor (1983) has It also shows that intervals of fixed prices claimed that it is possible for the monetary lasting months or years in some industries authorities to engineer a disinflation with can coexist with frequent small price no output loss. However, this result de- changes in other industries. It stresses that pends heavily on Taylor'sassumption that some industries are competitive, some are the effect of real demand on wage-setting monopolistic, and some industries combine decisions works through expected future monopolistic with competitive price real demand rather than past and current wars. Within industries, all firms do not Gordon: What Is New-Keynesian Economics? 1161

exhibit the same price behavior, and given firms maximize profit by setting their own firms do not even charge the same prices marginalcost equal to their own marginal to all customers. Heterogeneity is ram- revenue. They have no reason whatsoever pant, with hundreds of products common to care about nominal GNP unless it pro- in many industries, and many products vides useful information to supplement combine labor with hundreds or thousands what they can learn from observing their of purchased components. "local"cost and demand. There are many The time-series evidence shows a wide reasons for firms to expect their nominal variety of price adjustment patterns across marginalcosts and local demand to contain time and countries. The inertia (A) effect idiosyncratic elements that cause them to has become more prevalent since World evolve independently from nominal de- War II in every country but Japan. The mand. The most straightforwardargument, rate-of-change (ox)and level (y) effects were which is enough to make the case, is that remarkably similar before World War I and firms in a small open economy know that after World War II in most countries, but their costs are determined outside the na- exhibited sharp divergences in between. tional boundaries within which domestic The rate-of-change effect increased sharply nominal demand applies. This principle during and after World War I, while the generalizes to firms in large open econo- level effect virtually disappeared during mies, because we know that even under the interwar period in the U. S., U. K., and flexible exchange rates purchasing power Germany. The inertia-prone postwar U.S. parity does not hold over long periods, so is at one extreme, and Japan throughout costs of imports and domesticallyproduced most of its history is at the other extreme import substitutes can evolve indepen- of relative flexibility. dently of domestic aggregate demand. A convenient image for understanding The independence of cost and demand, the desirable direction of new-Keynesian and the input-output table approach, rep- theory is a small 1 x 1 box set next to a resent two separate components in the re- gigantic n X n matrix, where n is measured quired (but as yet missing) new-Keynesian in the thousands, if not the millions. The analysis that can come to grips with the small box represents the identical repre- industry, cross-time, and cross-country sentative agents of both new-classical mod- facts summarized here. The idea of inde- els and the canonical monopolistic compe- pendent cost and demand shocks seems tition model, with their i subscripts, and crucial to come to grips with the time-se- the practice of treating the macro economy ries evidence. Just as Lucas argued (1973) as identical to the representative agent that Argentinahad a more vertical Phillips with the subscripts removed. The gigantic curve because agents knew that aggregate matrix represents the real world, full of demand shocks dominated local shocks, so heterogenous firms enmeshed in a web of we can argue in parallel that the increase intricate supplier-demander relationships. in the rate-of-changecoefficient (ox)in Ta- This n X n matrix suggests two main ble 4 for the U. S., U. K., and Japanduring themes of the theoretical review in this pa- 1915-22 reflected a recognition by price per. setters that the increasing importance of First, the key to introducing theories of aggregate disturbances created a greater real rigidity as a source of nominal price than usual correlationbetween changes in stickiness is to find a good reason why we marginal costs and changes in aggregate do not observe nominal GNP indexation. demand. Similarly, the increase in persis- That reason is simple, and is at the heart tence (the A parameter) observed almost of all good microeconomics. Individual everywhere after World War II reflects a 1162 Journal of Economic Literature, Vol. XXVIII (September 1990) widespread belief that government full- local demand, and aggregatedemand is ad- employment policies and the end of the mitted as the fundamentalexplanation for standard created an upward drift in the lack of nominal demand indexation, the prices, leading to the expectation that mar- way is open to take seriously new-Keynes- ginal costs would have an upward drift and ian research on real rigidities in the labor would no longer be a stationary process. market. Work on union behavior and on The input-output component is comple- nominal contracting in the labor market mentary to the independent shocks idea, does not appear promising, in light of the and helps to explain why firms do not sim- similarityof the oxand y coefficientsin most ply assume that marginal costs will move countries before World War I and after in parallelwith aggregatenominal demand: World War II. However, the efficiency Most firms do not know the identity of their wage model has strong persuasive power suppliers, their suppliers' suppliers, and as to why firms resist real wage cuts, and so on, because the input-output table is the independence of shocks and input-out- so broad and so deep. The input-output put table explanations contribute the component of the proposed explanation is needed supplementaryexplanation of why required to grapple with the industry evi- real wage rigidity becomes translated into dence. Prices of corn and wheat on auction nominal wage rigidity. The other most marketsexhibit sharp daily swings, subject promising development in the labor mar- to administered limits. Prices of strawber- ket literature is the insider-outsider ap- ries exhibit frequent sharp weekly swings. proach, if only because the disenfranchise- Prices of many crude materialsexhibit fre- ment of outsiders holds up the best quent changes, both small and large. Yet available ray of hope that we have for un- prices of and many finished derstanding why the Phillips-curve level goods can remain unchangedfor more than (,y) effect disappeared in the U.S., U.K., a year. A unified explanationthat explains and Germaninterwar periods, and perhaps the degree of volatility of fixity for every in some European countries in the 1980s. product may be impossible to achieve, but Our perspective that emphasizes in- the basic idea that crude materialsare rela- dependent shocks and the input-output tively volatile and finished goods relatively approach reinforces the view that coordi- fixed seems compatible with the input-out- nation failures are the essence of macro- put approach which stresses the number economic in new Keynesian of steps and number of purchased compo- models. Should the government attempt nents that are mixed together with labor to intervene to provide the missing co- input in each . The input-output ordination of microeconomic wage and approachalso leaves open a role for a the- price decisions, or should its activities be ory of real wage rigidity, once it is admitted limited to the traditionalKeynesian use of that nominal GNP indexation is unlikely. monetary and to manipulate The input-outputapproach emphasizes the aggregate demand directly? Clearly, tradi- time lags in transmittingnews of cost and tional forms of internalizationthrough tax demand changes back and forth within the and subsidy policy are infeasible in light input-output table. However, to explain of pervasive heterogeneity among products why prices do not change by small amounts and decision makings in the millions; to every day, this approachneeds to be sup- go in this direction would mean slipping plemented with a plausible mixture of into the quagmirefrom which Eastern Eu- time-dependent and state-dependent costs rope is trying to emerge. Even mandatory of daily price changes. indexation to domestic nominal demand Once the independence of local costs, may be suboptimal in many countries Gordon: What Is New-Keynesian Economics? 1163 where an importantcomponent of nominal features will be considered the norm, marginal cost is set in foreign rather than some aberrant or exotic flower. and responds more to foreign than to do- Rather than thinking of basic aspects of mestic aggregate demand. This shifts the transaction and capital-market technology ultimate weapon for fighting business cy- as imperfections, perhaps we could all start cles back to the traditionalinstrument, ag- recognizing that these features are part of gregate demand policy, but not in the form the way that markets function. of any old-fashionedKeynesian bias in favor But these suggestions represent only the of fiscal policy. If prices respond slowly to beginning of a needed research program. fluctuationsin nominal GNP growth, then At the truly micro-micro level of relations the optimal objective of stabilizationpolicy between individual firms and customers, should be to stabilize the growth rate of imperfections go far beyond anything that nominal GNP growth. Whether and how the independence of shocks, input-output, this can be achieved is beyond the scope or efficiency wage approaches can explain of this paper.55 by themselves. The evidence presented by Some commentators (particularly Carlton that firms charge different prices Blanchard 1987a) have lamented that, far to different customers for the same prod- from being a set of facts looking for a the- uct, and apply nonprice allocation rules ory, the new-Keynesian paradigm suffers differently across customers, opens up from too many unrelated theoretical expla- a whole new dimension of heterogeneity nations. Yet the essential features empha- that future theorists will need to consider. sized here, the independence of shocks, The ultimate merger of the new empiri- and the input-output table, embody a core cal industrial organization and the new- set of realistic microeconomic elements: A Keynesian macroeconomics (it is hoped not technology of transactions, heterogeneity by ) seems a long way off, of goods and factor inputs, imperfect com- but it is a worthy goal to support. petition, imperfect information, and im- APPENDIX A perfect capital markets. Unlike time-de- THE VARIETY OF HISTORICAL EXPERIENCE: REGRESSION pendent or place-dependent factors like METHODOLOGY AND ESTIMATES these essential are time- unions, features Specification of Regression Equations less and placeless. They lead us to expect The aim is to estimate the three parameters A, that the degree of price flexibility in the ox, and y in equation (9) in the text, which is repeated early nineteenth century would not be here for convenience: much than insofaras greater today, except Pt = XPt-1 + oXxt+ YQt + zt. (9) X the n n matrix was smaller, with fewer There may be some concern regarding the close re- steps from primary producer to final con- semblance of (9) to an identity obtained by rearrang- sumer, and indeed we find a basic similar- ing (3): ity within each country in the ot and y pa- Pt a X-Qt + ot-l (a) rameters before World War I and after Comparin,g (9) and (a), the former includes Pt-i and World War II. excludes Qt_j. Because inertia may be absent in some Recognition of the universality of these historical eras (A = 0), the difference between (9) and (a) boils down to the exclusion of Q,1. Thus if imperfectionsin economic life is overdue- (9) is a true structural equation, the identity (a) pro- perhaps a campaign can be started to vides the value of the missing variable, Qt-,. This change economic language so that these argument is more transparent when (9) is transformed to include Qt-1 but to exclude Qt:

5 Advantages, problems, and Pt = [1/(1 + y)][XPt-1 + techniques relevant + + + to the targeting of nominal GNP growth are discussed (o y)it YQt-i zt. (9') in Tobin (1983), Hall (1984), McCallum (1988), and If (9) is a structural relation, so is (9'). Given the R. Gordon (1985). values of the right-hand variables in (9'), two of which 1164 Journal of Economic Literature, Vol. XXVIII (September 1990)

are predetermined (Pt-i and Q0-,) and one of which in 1960 to the BLS index of average hourly earnings is endogenous (h), the role of the identity (a) is to in manufacturing. The real wage is this nominal wage determine output as Qt Qt-1 + t - Pt. In short, series divided by the GNP deflator. Data sources the identity shows how output must change, given are given in Appendix B. the structuralprice equation (9'). This just restates The use of output data in estimating (9') and (9") the basic point about Keynesian economics: If the requires a detrending procedure to define the x, q, current price is predetermined by an equation like and Q variables. Significant variations in population (9'), then the current output level Qtis determined and productivity growth over the past century pre- as a residual. vent the use of a single trend and require the choice The main estimation problem is not the fact that of benchmark years to separate multiple piecewise there is an identity linking some of the variables in log-linear output trends. The choice of the wrong (9'), but ratherthe endogeneity of it, which we have benchmark years would introduce measurement er- discussed above in the context of policy feedback. ror into all three of these variables. To avoid the Because the essence of the problem is policy feed- possible criticism that benchmark years might have back, there can be no escape by replacing nomi- been selected to support or refute a particular hy- nal GNP by the money supply, or by using money pothesis, all are copied from previous research di- as an instrument for nominal GNP. And alternative rected at other issues.' The main control for supply versons with real GNP or unemployment are also shocks is a set of dummy variables to proxy the effects subject to bias if policy feedback is not complete, of government intervention both in the form of price as illustrated in Table 1. Our solution, which is controls (as during World War II) and intervention to bracket the ot parameter by estimating alterna- to raise prices and wages, as during the National tive versions of (9') with it and qt as alternative Recovery Act period in the U.S. Great Depression. explanatory variables, seems to be the best alter- Also for the U.S. we include a variable to measure native. the effect on aggregate inflation of changes in the To do this, we provide a pair of estimates for each relative prices of and energy.2 The specific dependent variable (price change, nominal wage values of the supply-shock dummy variables are given change, and real wage change for the U. S., and price in Appendix B. change for the U.K., France, Germany, and Japan). The key issue of changing cyclical responsiveness The specification for the first member of each pair can be addressed by two alternative methods. One is (9'). The specification for the second member of obvious way of providing information on parameter each pair is the transformationof (9') that results shifts would be to estimate separate versions of (9') when identity (a) is used to replace it by qt: and (9")for each major subperiod within the available

Pt = [1/(1 - 00]Xpt_I + (ox+ Y)qt+ YQt-i+ Zt]. (9) 1 For the postwar U. S., benchmark years are taken from my macroeconomics textbook (1990) and for the The values of the three parametersA, ox,and y can other four countries from Gordon (1988); for the pre- be easily unscrambled. If in (9') a, is the estimated World War II period, U.S. benchmarks are taken coefficient on it, a2 is the estimated coefficient on from (1989), for France, Germany, Pt-i, and a3 is the estimated coefficienton Qt-i, then and the U.K. from Solomos Solomou (1987), and for the parametersresulting from the estimation of (9') Japan from Gordon (1983). Inconsistency may result are y = [a3/(- a3)], ot = a, - y(l - a,), and K = from the use of benchmark years originally selected a2(1 + y). If in (9") b, is the estimated coefficient by varying criteria-peak output in some cases, aver- on qt, b2 is the estimated coefficient on Pt-i, and b3 age output in others, and the level of output consis- is the estimated coefficienton Qt-1, then the parame- tent with a particular unemployment rate in still oth- ters resulting from the estimation of (9") are t = ers. For the European countries, where the (b1 - b3)/(1 + bi - b3), y = b3(1 - a), and k = benchmark years before World War I are all peaks b2(1 - a (thus eliminating any positive values of Qt), the re- sulting Qt series is adjusted by subtracting its (nega- Data, Detrending, and Parameter Shifts tive) mean and converting the mean to zero. This results in a mix of positive and negative values. No Postwar data are taken from standard U.S. and such adjustments are carried out in the interwar of OECD sources, and data prior to World War II are postwar periods. based on Nathan Balke and R. Gordon (1989) for 2 larger number of such supply-shock variables the U.S., Charles Feinstein (1972) for the U.K., Ka- for the U.S. than for other countries may indicate zushi Ohkawaand Mihohei Shinohara(1979) for Ja- that supply shocks have been more important in the pan, and national sources as summarizedby Angus U.S., or they may simply indicate that I am more Maddison (1982) for France and Germany. Data for familiar with the history of the U. S. than of the other Japan, the U.K., and the U.S. measure nominal countries. However, the extra attention given to the GNP, real GNP, and the GNP deflator. Data for U.S. is largely due to the inclusion of wartime data Germanyand France prior to World War II measure for the U.S. but not for the other countries, where real GNP, the CPI, and a hybrid concept of nominal the years of World War II and its aftermath are ex- GNP equal to the CPI times real GNP. The nominal cluded for all four of the other countries, while World wage equations for the U. S. are based on Rees' data War I and its aftermath are excluded for France and on average hourly earnings in manufacturinglinked Germany. Gordon: What Is New-Keynesian Economics? 1165

TABLE A EQUATIONs EXPLAINING ANNUAL CHANGES IN THE GNP DEFLATOR, THE NOMINAL WAGE RATE, AND THE REAL WAGE RATE IN THE U.S., 1873-1987

Variable Price Nominal Wage Real Wage

(1) (2) (3) (4) (5) (6) Lagged inflation (Pt-1) Basic effect 0.23* 0.38* 0.32* 0.55* 0.09 0.18** Extra effects: 1915-22 0.38** 0.31 -0.07 1954-87 0.45* 0.66* 0.15 0.37** -0.31* -0.29** Excess nominal GNP growth (it) Basic effect 0.35* 0.44* 0.09** Extra effects: 1915-22 0.40* 0. 37* -0.04 Excess real GNP growth (4t) Basic effect 0.32* 0.51* 0.19* Extra effects: 1915-22 0.73* 0.89* 0.16 Detrended log output (Ot-1) Basic effect 0.22* 0.29* 0.30* 0.43* 0.08 0.14** Extra effects: 1930-53 -0.16* -0.21** -0.31* -0.41* -0.16** -0.20* Supply-shock variables (zt) World War I controls -4.97* -6.98** 2.00 -1.85 6.99* 5.13** NRA 7.21* 9.52** 13.12* 17.53* 5.91** 8.01* World War II controls -16.68* -20.19* -2.13 -8.87** 14.54* 11.32* Nixon controls -2.94 -3.60 -2.47 -3.81 0.46 -0.21 Food-energy effect 0.63 0.97 0.39 0.81 -0.24 -0.16 R2 0.85 0.52 0.83 0.58 0.36 0.43 S.E.E. 2.02 3.59 2.45 3.80 2.53 2.38 Durbin-Watson 2.10 2.09 2.08 2.11 2.23 2.15

Notes: Supply-shock variables are defined in Appendix B. * indicates statistically significant at 1 percent level, ** at 5 percent level.

data set. An alternative method, carried out previ- the U. K., France, Germany, and Japan. No years ously in Gordon (1983), involves estimating a single were omitted for the U.S. equation for the entire period for which data are available, and then searching for parameter shifts. Regression Results If additional variables are defined as the product of Table A addresses the issue of changing cyclical the three economic variables of interest (Pt-i, it or responsiveness of prices, nominal wage rates, and qt, and Q(t_) and "O, 1" dummy variables for each real wage rates in the U.S.3 Six columns of results subperiod, then the t ratios on the additional vari- are shown for the entire 1873-1987 sample period, ables provide estimates of the statistical significance with equations for price, nominal wage, and real wage of parameter shifts. In developing the results dis- changes presented in pairs. The first member of each played in Tables A and B, a search procedure was pair uses specification (9') in which nominal GNP followed in an attempt to locate parameter shifts dur- ing the following subperiods: first year through 1914, 3 Here the wage data are adjusted for the trend 1915-22, 1923-38 (1930-53 for the U.S.), and 1960- in productivity growth (using piecewise linear trends 86 (1954-87 for the U. S.). All of the statistically signif- between benchmarks), so that the dependent vari- icant parameter shifts are listed separately in Tables able in the columns labeled Nominal Wage is actually A and B. Because of severe declines in output during the change in trend unit labor cost, and in the col- wartime and postware recovery periods, the follow- umns labeled Real Wage is actually the change in ing years are omitted from the regression equations: labor's income share adjusted for cyclical fluctuations 1914-24 for France and Germany, and 1939-59 for in productivity. 1166 Journal of Economic Literature, Vol. XXVIII (September 1990)

TABLE B EQUATIONs EXPLAINING THE ANNUAL INFLATION RATE, FIVE COUNTRIES, 1873-1986

Country U. K. France Germany Japan (Years Omitted) 1939-59 1914-24, 1914-24, 1939-59 1939-59 1939-59

Lagged inflation (Pt-1) Basic effect 0.18* 0.22* -0.15** -0.42* 0.00 0.27** 0.11* 0.71* Extra effects: 1923-38 0.27* 0.69* -0.27** 1960-86 0.24* 0.72* 0.59* 1.41* 0.60* 0.70* Excess nominal GNP growth (it) Basic effect 0.58* 0.61* 0.72* 0.74* Extra effects: 1915-22 0.12** 0.17* 1923-38 -0.26** -0.28* 1960-86 -0.27** Excess real GNP growth (4t) Basic effect 0.30* 0.14 0.01 - 0.37 Extra effects: 1915-22 1.33* - 2.91 1923-38 1.23* 0.72* 1960-86 - 0.38 Detrended log output ((t-,) Basic effect 0.26* 0.36* 0.36* 0.03 0.27* 0.11* 0.28* 0.34* Extra effects: 1915-22 0.83* 1923-38 -0.18* -0.27* 0.88* -0.21* Supply-shock variables (zt) U.K. World War I -10.16* -23.48* U.K. 1972-73 controls -6.37* -10.24* U.K. 1976-77 Social Contract -4.38* -7.55* France Poincare 10.48* 26.69* France Popular Front 16.76* 44.37* Hitler controls -3.82** -5.97 Japan oil shock 6.63* 13.54** ft2 0.96 0.85 0.94 0.82 0.86 0.60 0.95 0.45 S.E.E. 1.28 2.46 1.51 2.65 1.27 2.15 1.72 5.44 Durbin-Watson 1.99 2.03 1.88 2.35 1.73 1.88 1.55 1.62

Notes: Sample period for U.K. begins in 1958, and for Japan begins in 1888. Supply-shock variables are defined in Appendix B. * Indicates statistically significant at 1 percent level, ** at 5 percent level.

change (i,) appears and the second member uses (9") APPENDIX B: DATA APPENDIX in which real GNP change (4,) appears as an alterna- tive. The separate lines within each group of explana- United States tory variables report several extra effects, that is, the coefficients on the product of the variable con- GNP, deflator, and food-energy effect 1929-87: cerned and a 0,1 dummy variable for the period Output and prices from National Income and Product shown. A parallel presentation of results for price- Accounts, Tables 1. 1 and 7.4., U.S. Department of change equations only is provided in Table B for . Food-energy effect (1959-87 only) is the the other four countries. The parameters are un- difference between the growth rates of the fixed- scrambled in Tables 3 and 4 of the text, and the weight consumption deflator and the fixed-weight results are interpreted in Sections III.B and III.C. deflator for consumption expenditures net of food Gordon: What Is New-Keynesian Economics? 1167

and energy, from National Income and Product Output trend The output trend is calculated as a Accounts, Table 7.1. 1869-1928: Balke and Gordon log-linear trend between the following benchmark (1989, Table 10). years: 1885, 1890, 1903, 1914, 1919, 1929, 1938, 1953, 1961, 1972, 1979, and 1987. Nominal wage rate 1960-87: BLS average hourly earnings in manufacturing, Economic Report of the Dummy variable: Japan Oil Shock: 1974 = 1.0. President (1989, Table B-44). 1888-1959: Rees' series on real CPI-deflated average hourly earnings REFERENCES in manufacturing, series B-70 in Long-term , 1860-1970, U.S. Department of AKERLOF, GEORGE AND YELLEN, JANET. "A Near-ra- Commerce, 1973, divided by CPI, series B-69. tional Model of the Business Cycle with Wage and Price Inertia," Quart. J. Econ., 1985 suppl., Output trend The output trend is calculated as a 100(5), pp. 823-38. log-linear trend between the benchmark years 1869, ALCHIAN, ARMEN. "Information Costs, Pricing, and 1873, 1884, 1891, 1900, 1910, 1924, and the quarterly Unemployment," Western Econ. J., June data for the quarters 1949:Q1, 1954:Q1, 1957:Q3, 1969, 7(2), pp. 109-28. 1963:Q3, 1970:Q2, 1974:Q2, 1979:Q3, and 1987:Q3. ALLEN, STEVEN G. "Changes in the Cyclical Sensitiv- For further details, see Gordon (1990, Appendix C). ity of Wages in the United States, 1891-1987," working paper without number, North Carolina Dummyvariables World War I: 1918 = 1.0, 1919- State U., Aug. 1989. 20 = 0.5. NRA: 1933-34 = 0.5, 1935-36 = -0.5. ALOGOSKOUFIS, GEORGE AND SMITH, RON. "The Phil- World War II: 1943-44 = 0.5, 1946-47 = -0.5. lips Curve and the : Some Historical Nixon: 1972-73 = 0.5, 1974 = -0.3, 1975 = -0.7. Evidence," working paper without number, Birk- beck College, , Mar. 1989. France and Germany AZARIADIS, COSTAS. "Implicit Contracts and Under- employment Equilibria," J. Polit. Econ., Dec. GNP and prices 1960-86: Real GNP and deflator 1975, 83(6), pp. 1183-1202. from OECD Statistics Paris (1988). 1870-1959: Real BACKUS, DAVID K. AND KEHOE, PATRICK J. "Interna- GNP and CPI from Maddison A (1982, Appendices tional Evidence on the Historical Properties of and E). Business Cycles," working paper without number or place, May 1988. Output trend The output trend is calculated as a BAILY, MARTIN NEIL. "Wages and Unemployment log-linear trend between the following benchmark under Uncertain Demand," Rev. Econ. Stud., Jan. years. For France: 1870, 1875, 1882, 1892, 1899, 1974, 41(1), pp. 37-50. 1904, 1912, 1924, 1939, 1951, 1964, 1972, and 1979. BALKE, NATHAN AND GORDON, ROBERT J. "The Esti- For Germany: 1870, 1874, 1884, 1890, 1900, 1907, mation of Prewar Gross National Product: Method- 1913, 1925, 1928, 1938, 1952, 1961, 1972, and 1979. ology and New Evidence," J. Polit. Econ., Feb. For both countries, growth in 1979-86 is calculated 1989, 97(1), pp. 38-92. by applying the 1972-79 growth in the capital-output BALL, LAURENCE; MANKIW, N. GREGORY AND ROMER, ratio to the observed growth of the capital stock, as DAVID. "The and the in Schultze (1987). Output-Inflation Trade-off," Brookings Pap. Econ. Dummy variables France Poincare: 1926 = 1.0. Act., 1988, (1), pp. 1-65. France Popular Front: 1936-38 = 0.33. BALL, LAURENCE AND ROMER, DAVID. "Real Rigidi- Hitler controls: 1937-38 = 0.5. ties and the Non-," NBER Working Paper no. 2476, Oct. 1987. United Kingdom . "Are Prices Too Sticky?" Quart. J. Econ., Aug. 1989a, 104(3), pp. 507-24. GNP and deflator 1960-86: OECD Statistics Paris . "The Equilibrium and Optimal Timing of (1988). 1870-1959: Feinstein (1972). Price Changes," Rev. Econ. Stud., 1989b, 56(2), pp. 179-98. Output trend The output trend is calculated as a BARRO, ROBERT J. "A Theory of Monopolistic Price log-linear trend between the following benchmark Adjustment," Rev. Econ. Stud., Jan. 1972, 39(1), years: 1856, 1865, 1873, 1882, 1889, 1907, 1913, pp. 17-26. 1920, 1940, 1951, 1961, 1972, 1979, and 1987. . "Unanticipated Money Growth and Unem- ployment in the United States," Amer. Econ. Rev., Dummy variables U.K. World War I: 1915-18 = Mar. 1977a, 67(2), pp. 101-15. 0.25, 1919-20 = -0.5. U.K. 1972-73 Controls: . "Long-term Contracting, Sticky Prices, and 1972-73 = 0.5, 1974-75 = -0.5. U.K. 1976-77 ," J. Monet. Econ., July 1977b, Social Contract: 1976 = 1.0, 1980 = -1.0. 3(3), pp. 305-16. . "Unanticipated Money, Output, and the Japan Price Level in the United States," J. Polit. Econ., Aug. 1978, 86(4), pp. 549-80. GNP and deflator 1960-86: OECD Statistics Paris . "Second Thoughts on Keynesian Econom- (1988). 1870-1940: Kazushi Ohkawa and Mihohei ics," Amer. Econ. Rev., May 1979, 69(2), pp. 54- Shinohara (1979), Tables A9 and A50. 59. 1168 Journal of Economic Literature, Vol. XXVIII (September 1990)

BARRO, ROBERT J. AND GROSSMAN, HERSCHEL. "A BOWLES, SAMUEL; GORDON, DAVID M. AND WEISS- General Disequilibrium Model of Income and Em- KOPF, THOMAS. Beyond the waste land. NY: Dou- ployment," Amer. Econ. Rev., Mar. 1971, 61(1), bleday, 1983. pp. 82-93. BRESNAHAN, TIMOTHY F. "Industries with Market . Money, employment, and inflation. Cam- Power," in Handbook of industrial organization. bridge: Cambridge U. Press, 1976. Eds.: RICHARD SCHMALENSEEAND ROBERT WILLIG. BARRO, ROBERT J. AND RUSH, MARK. "Unanticipated Amsterdam: North-Holland, 1989, ch. 17, pp. Money and Economic Activity," in Rational expec- 1011-57. tations and . Ed.: STANLEY BRUNO, MICHAEL. "Theoretical Developments in the FISCHER. Chicago: U. of Chicago Press for NBER, Light of Macroeconomic Policy and Empirical Re- 1980, pp. 23-48, 72-73. search," NBER Working Paper No. 2757, Nov. BARSKY, ROBERT AND SOLON, GARY. "Real Wages over 1988. the Business Cycle," NBER Working Paper no. CAPLIN, ANDREW S. AND LEAHY, JOHN. "State-depen- 2888, Mar. 1989. dent Pricing and the Dynamics of Money and Out- BENASSY, JEAN-PASCAL. "Neo-Keynesian Disequilib- put," working paper without number or place, rium Theory in a Monetary Economy," Rev. Econ. Oct., 1989. Stud., Oct. 1975, 42(4), pp. 503-23. CAPLIN, ANDREW S. AND SPULBER, DANIEL F. "Menu . The economicsof marketdisequilibrium. NY: Costs and the Neutrality of Money," Quart. J. Academic Press, 1982. Econ., Nov. 1987, 102(4), pp. 703-25. BERNDT, ERNST R. AND WOOD, DAVID 0. "Engineer- CARLTON, DENNIS W. "The Rigidity of Prices," Amer. ing and Econometric Interpretations of Energy- Econ. Rev., Sept. 1986, 76(4), pp. 637-58. Capital Complementarity," Amer. Econ. Rev., . "The Theory and the Facts of How Markets June 1979, 69(3), pp. 342-54. Clear: Is Industrial Organization Valuable for Un- BERTOLLA, GIUSEPPE AND CABALLERO, RICARDO J. derstanding Macroeconomics?" in Handbook of "Kinked Adjustment Costs and Aggregate Dynam- industrialorganization, Vol. I. Eds.: R. SCHMALEN- ics," paper presented at Fifth Annual NBER Con- SEE AND R. D. WILLIG. Amsterdam: North-Hol- ference on Macroeconomics, Mar. 9-10, 1990. land, 1989a, pp. 909-46. BLANCHARD, OLIVIER J. "Price Asynchronization and . "The Theory of Allocation and Its Implica- Price Level Inertia," in Inflation, , and indexa- tions for and Industrial Structure." Un- tion. Eds.: RUDIGER DORNBUSCH AND MARIO HEN- numbered working paper, U. of Chicago Graduate RIQUE SIMONSEN. Cambridge: MIT Press, 1983, pp. School of Business, May 1989b. 3-24. CECCHETrTI, STEPHEN. "The Frequency of Price Ad- . "The Lucas Critique and the Volcker Defla- justment: A Study of the Newsstand Prices of Mag- tion," Amer. Econ. Rev., May 1984, 74(2), pp. 211- azines, 1953 to 1979,"J. ,1986, 31(3), 15. pp. 255-74. . "The Wage-Price Spiral," Quart. J. Econ., CLOWER, ROBERT W. "The Keynesian Counterrevo- Aug. 1986, 101(3), pp. 543-65. lution: A Theoretical Appraisal," in The theory of "Why Does Money Affect Output? A Sur- interest rates. Eds.: F. H. HAHN AND F. BRECH- vey, NBER Working Paper No. 2285, June 1987a. LING. London: Macmillan, 1965, pp. 103-25. (Handbook of . Eds.: B. COE, DAVID T. "Hysteresis and Insider-Outsider In- FRIEDMAN AND F. HAHN. North-Holland, June fluences on Industry Wages," working paper with- 1990.) out number or place, Mar. 1989. . "Aggregate and Individual Price Adjust- COOPER, RUSSELL. "Predetermined Wages and Prices ment," Brookings Pap. Econ., Act., 1987b, (1), and the Impact of Expansionary Government Pol- pp. 57-122. icy," working paper without number or place, BLANCHARD, OLIVIER J. AND FISCHER, STANLEY. Lec- Sept. 1989. tures on macroeconomics. Cambridge: MIT Press, COOPER, RUSSELL AND JOHN, ANDREW. "Coordinating 1989. Coordination Failures in Keynesian Models, BLANCHARD, OLIVIER J. AND KIYOTAKI, NOBUHIRO. Quart. J. Econ., Aug. 1988, 100(3), pp. 441-63. "Monopolistic Competition and the Effects of Ag- CROSS, ROD, ed. Unemployment, hysteresis, and the gregate Demand," Amer. Econ. Rev., Sept. 1987, natural rate hypothesis. Oxford and NY: Basil 77(4), pp. 647-66. Blackwell, 1988. BLANCHARD, OLIVIER J. AND SUMMERS, LAWRENCE DELONG, J. BRADFORD AND SUMMERS, LAWRENCE H. H. "Hysteresis and the European Unemployment "The Changing Cyclical Variability of ," NBER MacroeconomicsAnnual, 1986, Activity in the United States," in The American Cambridge: MIT Press, pp. 15-78. business cycle: Continuity and change. Ed.: ROB- BLINDER, ALAN S. "The Fall and Rise of Keynesian ERT J. GORDON. Chicago: U. of Chicago Press for Economics," Econ. Rec., Dec. 1988, 64, pp. 278- NBER, 1986, pp. 679-734. 94. DIAMOND, PETER A. "Aggregate Demand Manage- BLOMSTROM, MAGNUS; LIPSEY, ROBERT E. AND OHLS- ment in Search Equilibrium," J. Polit. Econ., Oct. SON, LENNART. "What Do Rich Countries Trade 1982, 90(5), pp. 881-94. with Each Other? R & D and the Composition of . A search equilibrium approach to the micro U.S. and Swedish Trade," NBER Working Paper foundations of macroeconomics.Cambridge: MIT No. 3140, Oct. 1989. Press, 1984. Gordon: What Is New-Keynesian Economics? 1169

DICKENS, WILLIAM T. AND LANG, KEVIN. "A Test of justment,"J. Econ. Lit, June 1981, 19(2), pp. 493- Dual Labor Market Theory," Amer. Econ. Rev., 530. Sept. 1985, 75(4), pp. 792-805. . "Why U.S. Wage and Employment Behavior DOERINGER, PETER B. AND PIORE, MICHAEL J. Inter- Differs from that in Britain and Japan," Econ. J., nal labor marketsand manpoweranalysis. Lexing- Mar. 1982a, 92(365), pp. 13-44. ton, MA: D.C. Heath and , 1971. . "Price Inertia and Policy Ineffectiveness in DUNLOP, JOHN T. "The Movements of Real and the United States, 1890-1980," J. Polit. Econ., Money Wage Rates," Econ. J., Sept. 1938, 48(191), Dec. 1982b, 90(6), pp. 1087-1117. pp. 413-34. . "A Century of Evidence on Wage and Price . Wage determination under trade unions. Stickiness in the United States, the United King- NY: Macmillan, 1944. dom, and Japan," in Macroeconomics, prices, and DURLAUF, STEVEN N. "Output Persistence, Eco- quantities. Ed.: . Washington, DC: nomic Structure, and the Choice of Stabilization , 1983, pp. 85-121. Policy," Brookings Pap. Econ. Act., 1989, 20(2), . "The Conduct of Domestic Monetary Pol- pp. 69-116. icy," in Monetarypolicy in our times. Eds.: ALBERT ECKSTEIN, OrrO AND BRINNER, ROGER. The inflation ANDO ET AL. Cambridge: MIT Press, 1985, pp. process in the United States. A Study Prepared 45-81. for the use of the Joint Economic Committee, 92 . "Wage Gaps vs. Output Gaps: Is There a Cong. 2 session, 1972. Common Story for All of Europe?" in Macro and .Economic report of the President 1989. micro policies for more growth and employment: Washington, DC: U.S. GPO, 1989. Keil symposium. Ed.: . Tubin- FEINSTEIN, C. H. National income expenditure and gen: J. C. B. Mohr [Paul Siebeck] 1988, pp. 97- output of the United Kingdom, 1855-1965. Cam- 151. bridge: Cambridge U. Press, 1972. . Macroeconomics, 5th ed. Glenview, IL: FISCHER, STANLEY. "Long-term Contracts, Rational Scott, Foresman, 1990. Expectations, and the Optimal Money Supply GORDON, ROBERT J. AND KING, STEPHEN R. "The Out- Rule,"J. Polit. Econ., Feb. 1977a, 85(1), pp. 191- put Cost of Disinflation in Traditional and Vector 205. Autoregressive Models," Brookings Pap. Econ. "Wage Indexation and Macroeconomic Sta- Act., 1982, 13(1), pp. 205-42. bility,"in Stabilizationof the domesticand interna- GORDON, ROBERT J. AND WILCOX, JAMES. "Monetarist tional economy. Eds.: AND ALLAN Interpretations of the Great Depression: An Evalu- MELTZER. Carnegie-RochesterConference Series ation and Critique," in The Great Depression re- on , vol. 5. Amsterdam: North-Hol- visited. Ed.: KARL BRUNNER. : Martinus land, 1977b, pp. 107-47. Nijhoff, 1981, pp. 49-107. . Indexing, inflation, and economic policy. GRAY, Jo ANNA. "Wage Indexation: A Macroeconomic Cambridge: MIT Press, 1986. Approach," J. Monet. Econ., Apr. 1976, 2(2), pp. . "Recent Developments in Macroeconom- 221-35. ics," Econ. J., June 1988, 98(391), pp. 294- HALL, ROBERT E. "Employment Fluctuations and 339. Wage Rigidity," Brookings Pap. Econ. Act., 1980, FRIEDMAN, MILTON. "The Role of Monetary Policy," (1), pp. 91-123. Amer. Econ. Rev., Mar. 1968, 58(1), pp. 1-17. "Monetary Policy with an Elastic Price Stan- FRIEDMAN, MILTON AND SCHWARTZ, ANNA. A mone- dard,"in Price stability and public policy. Federal tary history of the United States, 1867-1960. Reserve Bank of City, 1984, pp. 137-59. Princeton: Princeton U. Press for NBER, 1963. HOwIrr, PETER. "The Keynesian Recovery," Can. GORDON, DAVID M.; EDWARDS, RICHARD AND REICH, J. Econ., Nov. 1986, 19(4), pp. 626-41. MICHAEL. Segmentedwork, divided workers: The KASHYAP, ANIL. "Sticky Prices: New Evidence from historical transformation of labor in the United Catalogs." Finance and Economics Discus- States. NY: Cambridge U. Press, 1982. sion Series Working Paper No. 112, Federal Re- GORDON, DONALD F. "A Neo-Classical Theory of serve Board, Washington, DC, Mar. 1990. Keynesian Unemployment," Econ. Inquiry, Dec. KATZ, LAWRENCE F. "Efficiency Wage Theories: A 1974, 12(4), pp. 431-59. Partial ," NBER macroeconomics an- GORDON, ROBERT J. "Wage-Price Controls and the nual, 1986, 1, Cambridge: MIT Press, pp. 235- Shifting Phillips Curve," Brookings Pap. Econ. 75. Act., 1972, (2), pp. 385-421. . "Some Recent Developments in Labor Eco- . "The Impact of Aggregate Demand on nomics and Their Implications for Macroeconom- Prices," Brookings Pap. Econ. Act., 1975, (3), pp. ics,"J. Money, Credit, Banking, Aug. 1988, 20(3), 613-62. pp. 507-22. . "Can the Inflation of the 1970s Be Ex- KATZ, LAWRENCE F. AND SUMMERS, LAWRENCE H. plained?" Brookings Pap. Econ. Act., 1977, 8(1), "Industry Rents: Evidence and Implications," pp. 253-77. BrookingsPap. Econ. Act.: Microeconomics1989, . "A Consistent Characterization of a Near- 1989, 51, pp. 209-90. Century of Price Behavior," Amer. Econ. Rev., KEYNES, JOHN M. "Relative Movements of Real May 1980, 70(2), pp. 243-49. Wages and Output," Econ. J., Mar. 1939, 49(193), . "Output Fluctuations and Gradual Price Ad- pp. 34-51. 1170 Journal of Economic Literature, Vol. XXVIII (September 1990)

KYDLAND, FINN E. AND PREscorr, EDWARD C. "Time A Very Quick Refresher Course," J. Money, to Build and Aggregate Fluctuations," Econo- Credit, Banking, Aug. 1988, 20(3, part 2), pp. 436- metrica, Nov. 1982, 50(6), pp. 1345-70. 49. LAIDLER, D. E. W. Essays on money and inflation. . "Real Business Cycles: A New Keynesian Manchester, U.K.: Manchester U. Press, 1975. Perspective,"J. Econ. Perspectives, Summer 1989, LAIDLER, D. E. W. AND PARKIN, J. MICHAEL. "Infla- 3(3), pp. 79-90. tion-A Survey," Econ. J., Dec. 1975, 85(340), MARSHALL,ALFRED. Principles of economics. Lon- pp. 741-809. don: Macmillan, 1920. LAYARD, RICHARD AND NICKELL, STEPHEN. "The La- MCCALLUM,BSNNETrT. "On 'Real' and 'Sticky-Price' bour Market,"in The performance of the British Theories of the Business Cycle," J. Money, Credit, economy.Eds.: RUDIGER DORNBUSCH AND RICHARD Banking, Nov. 1986, 18(4), pp. 398-414. LAYARD. Oxford: Clarendon Press, 1987, pp. 131- . "Postwar Developments in Business Cycle 79. Theory: A Moderately Classical Perspective," J. LEIBENSTEIN, HARVEY. "The Theory of Underemploy- Money, Credit Banking, Aug. 1988, 20(3, Part 2), ment in Backward Economies," J. Polit. Econ., pp. 459-71. Apr. 1957, 65(2), pp. 91-103. . "Real Business Cycle Theory," in Modern LEIJONHUFVUD, AXEL. On Keynesianeconomics and Business Cycle Theory. Ed.: ROBERT J. BARRO. the economics of Keynes. NY: Oxford U. Press, Cambridge: Harvard U. Press, 1989, pp. 16-50. 1968. McDONALD, IAN AND SOLOW, ROBERT M. "Wage Bar- . Information and coordination: Essays in gaining and Employment," Amer. Econ. Rev., macroeconomic theory. NY: Oxford U. Press, 1981. Dec. 1981, 71(5), pp. 896-908. LINDBECK, ASSAR. "Remaining Puzzles and Ne- . "Wages and Employment in a Segmented glected Issues in Macroeconomics," Institute for Labor Market," Quart. J. Econ., Nov. 1985, International Economic Studies (University of 100(4), pp. 1115-41. ), Seminar Paper No. 424, Dec. 1988. MEANS, GARDINERC. "Industrial Prices and Their LINDBECK, ASSAR AND SNOWER, DENNIS. "Wage Set- Relative Inflexibility," U.S. Senate Document 13, ting, Unemployment, and Insider-Outsider Rela- 74th Congress, 1st session, Washington, DC, tions," Amer. Econ. Rev., May 1986, 76(2), pp. 1935. 235-39. MILLS, FREDERICK C. The behavior of prices. NBER . The Insider-outsider theory of employment General Series, no. 11. NY: Arno Press, 1927. and unemployment. Cambridge: MIT Press, 1988. MITCHELL, DANIEL J. B. "Wage Flexibility: Then and . "Remaining Puzzles and Neglected Issues Now," Ind. Relat., Spring 1985, 24, pp. 266-79. in Macroeconomics," Scand. J. Econ., 1989, 91(2), MORTENSEN, DALE T. "A Theory of Wage and Em- pp. 495-516. ployment Dynamics," in PHELPS, 1970, pp. 167- LIPSEY, RICHARDG. "The Relation between Unem- 211. ployment and the Rate of Change of Money Wage . "Job Search, the Duration of Unemploy- Rates in the United Kingdom, 1862-1957: A Fur- ment, and the Phillips Curve," Amer. Econ. Rev., ther Analysis," Economica, N.S., Feb. 1960, 27, Dec. 1970b, 60(5), pp. 847-62. pp. 1-31. National income and product accounts of the United Long-term economic growth, 1860-1970. Washing- States, 1929-82. Washington, DC: U.S. Dept. of ton, DC: U.S. Dept. of Commerce, 1973. Commerce, 1986. (Updated beyond 1982 from LUCAS, ROBERT E., JR. "Expectations and the Neu- Surv. Curr. Bus., various issues) trality of Money," J. Econ. Theory, Apr. 1972, NORDHAUS, WILLIAM D. "Recent Developments in 4(2), pp. 103-24. Price Dynamics," in The econometrics of price de- . "Some International Evidence on Output- termination. Ed.: Orro ECKSTEIN. Washington, Inflation Tradeoffs," Amer. Econ. Rev., June 1973, DC: Board of Governors of the 63(3), pp. 326-34. System, 1972, pp. 16-49. . "Econometric Policy Evaluation: A Cri- OECD Statistics. Paris, 1988. (Data obtained from tique," in The Phillips curve and labor markets. data diskette.) Eds.: KARL BRUNNER AND ALLAN MELTZER. Am- OHKAWA, KAZUSHI, AND SHINOHARA, MIHOHEI, eds. sterdam: North-Holland, Carnegie-Rochester Se- Patterns of Japanese : A ries on Public Policy, a supplementary series to quantitative appraisal. New Haven: Yale U. Press, J. Monet. Econ., Jan. 1976, 1, pp. 19-46. 1979. LUCAS, ROBERT E., JR. AND SARGENT, THOMAS J. Af- OKUN, ARTHUR M. "Inflation: Its Mechanics and Wel- ter Keynesian Macroeconomics," in After the Phil- fare Cost," Brookings Pap. Econ. Act., 1975, (2), lips curve: Persistence of high inflation and high pp. 351-401. unemployment. Boston: Federal Reserve Bank of . Prices and quantities: A macroeconomic Boston, 1978, pp. 4972. analysis. Washington, DC: Brookings Institution, MADDISON, ANGUS. Phases of capitalist development. 1981. Oxford: Oxford U. Press, 1982. PARKIN, MICHAEL. Macroeconomics. Scarborough, MANKIW, N. GREGORY. "Small Menu Costs and Large Ontario: Prentice-Hall, 1984. Business Cycles: A ," PATINKIN, DON. Money, interest, and prices, 2nd ed. Quart. J. Econ., May 1985, 100(2), pp. 529-38. NY: Harper & Row, 1965. . "Recent Developments in Macroeconomics: PHELPS, EDMUND S. Microeconomic foundations of Gordon: What Is New-Keynesian Economics? 1171

employment and inflation theory. NY: Norton, Unemployment in Europe," in Barriers to - 1970. pean growth: A transatlantic view. Eds.: ROBERT .Inflation policy and unemploymenttheory: Z. LAWRENCE AND CHARLES L. SCHULIZE. Wash- The cost-benefit approach to monetary planning. ington, DC: Brookings Institution, 1987, pp. 230- NY: Norton, 1972. 91. . : An introductory text. NY: SHAPIRO, CARL AND STIGLITZ, JOSEPH. "Equilibrium W. W. Norton, 1985. Unemployment as a Discipline Device," Amer. PHELPS, EDMUND S. AND TAYLOR, JOHN B. "Stabiliz- Econ. Rev., June 1984, 64(3), pp. 433-44. ing Powers of Monetary Policy Under Rational Ex- SHESHINSKI, EYTAN AND WEISS, YORAM. "Inflation and pectations,"J. Polit. Econ., Feb. 1977, 85(1), pp. Costs of Price Adjustment," Rev. Econ. Stud., 163-90. June 1977, 44(2), pp. 287-303. PHELPS, EDMUND S. AND WINTER, SIDNEY G., JR. . "Optimum Pricing Policy under Stochastic "Optimal Price Policy Under Atomistic Competi- Inflation," Rev. Econ. Stud., July 1983, 50(3), pp. tion," in PHELPS, 1970, pp. 309-37. 513-29. PHILLIPS, A. W. "The Relation between Unemploy- SOLOMOU, SOLOMOS. Phasesof economicgrowth, ment and the Rate of Change of Money Wage Rates 1850-1973. Cambridge: Cambridge U. Press, in the United Kingdom, 1861-1957," Economica, 1987. Nov. 1958, 25, pp. 283-99. STIGLER, GEORGE AND KINDAHL, JAMES K. The behav- PLOSSER, CHARLES I. "Understanding Real Business ior of industrial prices. NY: Columbia U. Press Cycles,"J. Econ. Perspectives, Summer 1989, 3(3), for the NBER, 1970. pp. 51-77. STIGLITZ, JOSEPH. "Theories of Wage Rigidity," in PREscorr, EDWARD C. "Theory Ahead of Business Keynes'economic legacy: Contemporaryeconomic Cycle Measurement," Carnegie-Rochester Confer- theories. Eds.: JAMES L. BUTKIEWICZ, KENNETH J. ence Series on Public Policy, Autumn 1986, 25, KOFORD, AND JEFFREY B. MILLER. NY: Praeger, pp. 11-44. 1986, pp. 153-206. ROBERTS, JOHN M.; STOCKTON, DAVID J. AND STRUCK- SUMMERS, LAWRENCE H. "Relative Wages, Effi- MEYER, CHARLES S. "An Evaluation of the Sources ciency Wages, and Keynesian Unemployment," of Aggregate Price Rigidity," Working Paper 99, Amer. Econ. Rev., May 1988, 78(2), pp. 383- Economic Activity Section, Division of Research 88. and Statistics, Board of Governors of the Federal TARSHIS, LORIE. "Real Wages in the United States Reserve System, May 1989. and Great Britain," Can. J. Econ., Aug. 1938, 4(3), ROMER, CHRISTINA. "The Prewar Business Cycle Re- pp. 362-76. considered: New Estimates of Gross National Prod- . "Changes in Real and Money Wages," Econ. uct, 1869-1908,"J. Polit. Econ., Feb. 1989, 97(1), J., Mar. 1939, 49(193), pp. 150-54. pp. 1-37. TAYLOR, JOHN. "Aggregate Dynamics and Staggered ROSEN, SHERWIN. "Implicit Contracts: A Survey," J. Contracts," J. Polit. Econ., Feb. 1980, 88(1), pp. Econ. Lit., Sept. 1985, 23(3), pp. 1144-75. 1-24. " ROTEMBERG,JULIO. "The New Keynesian Microfoun- Union Wage Settlements," Amer. Econ. dations," NBER Macroeconomics Annual, 1987, 2, Rev., Dec. 1983, 73(5), pp. 981-93. pp. 69-104. . "Improvements in Macroeconomic Stability: SACHS, JEFFREY D. "The Changing Cyclical Behavior The Role of Wages and Prices," in The American of Wages and Prices: 1890-1976," Amer. Econ. business cycle: Continuity and change. Ed.: ROB- Rev., Mar. 1980, 70(1), pp. 78-90. ERT J. GORDON. Chicago: U. of Chicago Press for SARGENT, THOMAS J. "A Note on the 'Accelerationist' NBER, 1986, pp. 639-77. Controversy," J. Money, Credit, Banking, Aug. TOBIN, JAMES. "Monetary Policy: Rules, Targets, and 1971, 3(3), pp. 721-25. Shocks," J. Money, Credit, Banking, Nov. 1983, SCHULIZE, CHARLES. "Some Macro Foundations for 15(4), pp. 506-18. Micro Theory," Brookings Pap. Econ. Act., 1981, . "On the Theory of Macroeconomic Policy," (2), pp. 521-76. Cowles Foundation Discussion Paper no. 931, . Other times other places: Macroeconomic Dec. 1989. lessonsfrom U.S. and European history. Washing- WEISS, ANDREW. Efficiency wages: Models of unem- ton, DC: Brookings Institution, 1986. ployment, layoffs, and wage dispersion. Princeton, . "Real Wages, Real Wage Aspirations, and NJ: Princeton U. Press, 1990.