Potential Output and Productivity

GEORGE L. PERRY Brookings Institution

Potential Ou tput an 'roductivity

THE UNUSUAL TURMOIL in the U.S. economy during the 1970s has forced a reexaminationof many of its macroeconomiccharacteristics. In- flationhas been the most surprisingfeature of the period and the investi- gationof its causesand searchfor cureshave been intense.The worsening of the relationbetween inflation and unemploymenthas been a focus of that investigationand has been linked,in part,to the veryhigh unemployment ratesexperienced by youngermembers of the workforce and to the remarkablerise in participationrates of young personsand adultwomen. These same developmentsbear on the measurementof potential gross national product, an importantunderpinning of both macroeconomic analysisand economic policymaking. The 1974-75 recessionand its aftermathhave sparkedreestimates of potentialGNP that are substantiallylower than the previousofficial estimates and that anticipatea slower growth of potentialin coming years than previous official projections.The Council of Economic Advisers lowered the former estimate of 1976 potential by $58 billion (1972 prices), reducing the gap between potential and actual GNP to $99 billion.' One elementin this revisionis a changein the unemploymentrate at which potentialis definedfrom the traditional4 percentto a shifting rate that was 4.9 percentin 1976. This adjustmentis relatedto the higher unemploymentrates experiencedby youngerworkers and the declining Note: I am grateful to Lisa L. Blum for extensive research assistance, to Evelyn M. E. Taylor for typing, and to Andrea V. Mills for computational help. Edward F. Denison kindly provided me with data from his own forthcomingstudy. 1. Economic ReDortof the President.January 1977. PP. 52-56.

11 12 Brookings Papers on Economic Activity, 1:1977 proportionof adult males in the total work force. Another important element in the CEA revisionsis the observedweakness in productivity since 1973. Outputper hourin the (nonresidentialnonfarm) business sector declinedby 31/2percent in 1974, the firstdecline of any size in the post- war years. Despite a sharpincrease during the 1976 economicrecovery, productivitythat year was still only 21/4percent above the 1973 level. The CEA analysisassociates this weaknesswith an inadequaterate of capital accumulationand with the sharprise in energyprices that coincidedwith the recession.These factors also lead them to projecta potentialgrowth rate of only 31/2percent a year through1980.2 Data Resources,Inc., has publishedan analysisthat lowers the estimateof the presentgrowth rate of potentialeven further, to 3.35 percent.3 A differentchallenge to traditionalestimates of potential output has come fromthose who questionthe effect of unemploymenton laborforce participation.The sustainedgrowth of the labor force right throughthe 1974-75 recessionyears surprisedmany observers.If this recentstrength of participationrates is viewed as evidence of their insensitivityto unemployment,the traditionalmarkup of the laborforce of secondaryworkers that is used in calculatingpotential output is called into question. Withouta large induced rise in the work force between today's actual economyand its potential,the present GNP gap would be substantially smallerthan calculationssuch as Okun'slaw imply. Theoreticalreasons for questioningany directlink betweenunemployment and the size of the laborforce have been providedby economistswho look, instead,for a sub- stantialresponse of laborforce palticipationto wages and prices.Michael Wachterhas raised such doubts about the cyclicalbehavior of participationrates in his own empiricalestimates of laborforce behavior.4 This paperwill reevaluatethe evidenceon the recent and prospective behavior of potential output. The issues just raised will be examined againstthe backgroundof a previouslysuccessful approach to estimating potential.5The analysiswill startwith the growthin the laborforce, which

2. The analytic backgroundfor the CEA revisions is available in Peter K. Clark, "A New Estimate of Potential GNP" (Council of Economic Advisers, 1977; processed). 3. Roger Brinner, "Potential Growth to 1980," in Otto Eckstein and others, Economic Issues and Parametersof the Next 4 Years (Data Resources, Inc., 1977). 4. Michael L. Wachter, "A Labor Supply Model for SecondaryWorkers," Review of Economics and Statistics,vol. 54 (May 1972), pp. 141-51. 5. George L. Perry, "LaborForce Structure,Potential Output,and Productivity," BPE1, 3:1971, pp. 533-65. George L. Perry 13 is an importantbuilding block in estimatesof potenitialand is of con- siderableinterest in itself.

LaborForce Participation

The substantialdifferences in the labor force behavior of different demographicgroups make it essentialto analyzegrowth in the laborforce on a disaggregatedbasis. At a minimum,data are availableby age-sex categories;and with some sacrificein statisticalreliability and historical continuity,some furtherdisaggregations are possible.For a givenpopula- tion cell, it is useful to decomposelabor force growth into changes in the populationand changesin participationrates-the proportionof the populationin the workforce. Analysisof laborforce growththen focuses on participationrates of individualpopulation groups. Rising participationby teenagersand womenhas been the outstanding and surprisingfeature behind the rapid growthof the labor force in the past decade. Sociologicalhypotheses to explain this changingbehavior come readilyto mind:the greaterindependence of teenagers,the changing values and attitudesfostered by the women'smovement, the betterfamily planningmade possibleby superiorcontraceptives, all must be involved. More narrowlyeconomic hypotheses can be offeredas well, built around the inflationof the last decade and the change in real living standards comparedwith expectations. To analyzethe presentlevel of potentialoutput and its past growth,one needs estimatesof the cyclicalresponse of participationrates so that the laborinput available along the potentialpath can be measured.I startwith some conventionallyestimated equations for participationrates and then consider whether alternativehypotheses about the observed growth in participationcan improvethese calculations.

CYCLICAL VARIATIONS IN PARTICIPATION

Accordingto the conventionalview, participationrates for some demographicgroups vary both cyclically,with a largerfraction of the population in the labor force when unemploymentis low, and secularly,for reasonsthat are independent of the unemploymentrate. Lacking any good way to identifyand quantifythe factorsbehind secular changes, past re- searchhas generallyallowed for themby usingtime-trend variables. After 14 Brookings Papers on Economic Activity, 1:1977 some experimenting,the followingform of estimatingequation was used for eachof fourteendemographic groups in the presentanalysis:

(1) log (LIP) = Ao+ aju*+ a2U*(- 1) + a3T54+ a4T67, whereL is the total laborforce and P is population;u* and u* (-1 ) are this year'sand last year'sweighted unemployment rates, respectively; and T54 and T67 are time-trendvariables starting at 1 in 1954 and 1967, respectively,and risingby 1 each year.6The conceptof a weightedunem- ploymentrate as a measureof labormarket tightness is by now familiar.7 The formof the dependentvariable was chosen aftersome preliminary comparisonsbetween it and log (1 - L/P), the "nonparticipationrate," showedno persistentsuperiority for either.One workedslightly better for some groupsand slightlyworse for others.Using the nonparticipationrate requiredmore abrupttrend changes (since it makes trendsin participation rise more slowly with time when, in fact, they speededup for some groups). And, while a constanttrend growthrate in participationrates such as equation1 implies cannot go on forever,that only warnsagainst extrapolatingit too far into the future. The choice of a second time trend startingin 1967 was made after experimentingwith startingpoints in 1964 and 1970. Once again, some trendsfitted better for some populationgroups and worse for others.The trend startingin 1964 sometimesreduced the estimatedcyclical responsiveness of participationrates, presumablybecause more of the strong rise that occurred as unemploymentfell between 1964 and 1966 was attributedto trend.These equationsgenerally did not fit the data quite as well as the others.And 1964 predatedthe sociological changesthat are hypothesizedto lie behindthe stronglyrising participation trends of teenagersand younger women. Therewas less differencebetween the estimatesusing the time trends 6. The data on the labor force, employment, and unemployment used here and throughout the paper (except as noted) are official estimates and are adjusted for Census-year changes and the changes in definition introduced in 1967. They are further adjustedto an employment basis (largely an establishmentbasis) that corresponds to the output categories in the GNP accounts. The adjustmentsare described in the appendixto Perry, "LaborForce Structure."The necessary data for the adjustment to an employment basis correspondingto GNP were provided by Edward F. Denison from his worksheetsfor a forthcoming study. 7. The version in this paper uses the proportionsof differentdemographic groups in the labor force in 1966 together with the relative-wageand average-hourfactors of each group that were explained in George L. Perry, "ChangingLabor Markets and Inflation,"BPEA, 3:1970, pp. 411-41. George L. Perry 15 startingin 1967 and 1970, and also less a prioribasis for preferringone to the other.The appendixsummarizes the cyclicalresponse, time trends, and goodnessof fit of equationsfitted to each demographicgroup using these alternativetime trendsand some alternativeperiods of estimation. In general,using a time trendstarting in 1970 resultedin slightlygreater cyclical responsiveness.So did ending the estimationperiod in 1973, before the great recession. Since all the estimates show a very strong cyclicalresponsiveness in participation,I chose the most conservative,the equationsusing the trend startingin 1967 and estimatedfrom 1954 to 1976, whichincludes the periodof particularinterest for estimatingpoten- tialin thisstudy.8 Table 1 shows the estimatedequations for the participationrate for eachof thedemographic groups. Since the severalequation forms shown in the appendixwere compared before the form of equations shown in table 1 was selected, the basis for retainingcoefficients relies on all the estimatesshown in the appendix.All the unemploymentcoefficients in table 1 have a t statistic of at least 1.6; the t statistics were usually higher in equations using a time trend starting in 1970. Time-trend coefficientswere kept if theirt statisticswere greaterthan 1.5; for all but the over-65age group,they are at least 2.0 in the table 1 estimates. For both sexes in the groups aged 16 to 24 and for women in the groupsaged 25 to 44, a statisticallysignificant and quantitativelyimportant speedupin the trend of participationrates is found using the 1967 time dummy.For teenagersand womenaged 25 to 34, this accelerationis a strikinglylarge 3 to 4 percenta year (on a base of roughly0.5 for participationrates in these groups). Men 25 to 64 andwomen 45 and over show a statisticallysignificant slowing in participationrates using the 1967 time dummy.The net result of the time trendson participationrates over the last ten yearsis shownin the last columnof table 1. Sharplyrising trends are estimatedfor women of all age groupsbetween 16 and 44, with the averageannual rate of increaseranging between about 2?/z and 4 percent. Teenagemales also show a pronouncedupward trend, while men over 54 and women over 64 have steep downwardtrends. The trends for other demographicgroups are much more gradual.

8. Using the same form of the trend, as well as the dependentvariable, for all the demographic groups was a way of minimizing the effects of the data mining that would have gone with selecting a "best-fitting"equation for each group after experimentingwith so many alternatives. 0 0 C %0 00 CM 4qt < o 8 8 8 o o

~~~ ~ n 00 0 R

S O\ 0 o 0 0. AI 4 I I I I

.PQ 12, O 0 CD N

0 N 0 ? 00 0O O O

*83~ ~ I I 0D 0% 00q O .

Q~~~~~~r-OC m b~FtN?t Nstc 0~~~~~~~0

* ~ ~~~ CM 0~~~0 5 Cj .'- M 0t *0A

ti~~~~~~~-

00 asN0 X ?0 *1

o% o ; ;t t~~in oN tq 00I

0~~~~~~~`

~~~~~~~~~~~~Int 5-I ~ ~ ~ ~ ~ ~ ~ ~ C n I- n~ to~~' t- r NN

0 *.04 * * eq Itl~ ~

c Ono0 C)o% Co. O oC) Co

Cf. Co C4 '4 0 0 0

A4-

to ,0 eq~~~~~~~ 0 00 V' t~-. Cq en 10 en. S 00 0 0 0 0 0 0 tn~~~~

n e e~~~~I 0

O ~O O- ,O OO) o o

5 0 0 tnk ,cMc

O t5 ^ ' 5 5 O O OI O ? d ~~~~~~~~oo 0 n > C;O CD 0%- ~CD 4.'Oo CD-~* O8

O O O ~~O O) O O

n E .4.

'o C' c 4 O . Ca

o W N N \0 t 00 : : ce,<'0._.4 N c>l m t1-ay cl b c>l *.S'0* (10 (1) 8 1 I~~~8 t8 n 8 t o I> oOI N OIo z =v:~~>. (1 I* I. _~ eI% _ , Ic _. . 0 0%

x a s'

Ofe)t o .'

0% Q t f E t r [ E5;~'3-'"

tl csl n t m so Ja~~~~~~; 18 Brookings Papers on Economic Activity, 1:1977 Cyclical Response. The cyclical responsivenessof participationrates is summarizedin the next to last column of table 1, which shows the percentagechange in participation(or the laborforce) thatwould accom- pany a sustainedincrease of 1 percentagepoint in the weightedunemployment rate. Althoughthese estimatesare the most conservativeof those availablefrom the severalalternatives shown in appendixtable A-1, they do predicta strongercyclical response, particularly for recentyears, than many earlier studies. According to previous estimates summarizedby Okun, a one-point decline in unemploymentadded 0.6 percent to the laborforce." For 1976, the presentestimates indicate that an incrementin the laborforce of about0.9 percentaccompanies each point of unemployment. However,since the large incrementsto the work force are concein- tratedamong young workers and younger adult women-workers who, on the whole, receive relativelylow productivityweights in the subsequent estimatesof potentiallabor input -the potentialoutput calculations are not overlyinfluenced by this differencein estimatesof cyclical participation-rateeffects. If the labor force incrementswere only three-fourthsas large, thus bringingthem down to the averagecited by Okun, potential (weighted) labor input in 1976 would be only about 0.25 percentbelow whatis estimatedusing the table 1 equations. Part of the reason for the large participationeffects estimatedin the table 1 equationscomes from effectsfrom laggedunemployment that turn up in the currentequations. And part resultsfrom the increasedrelative importanceof women and teenagers in today's work force. The table below analyzesthe changingcyclical responsiveness of the aggregatelabor force (in termsof percentagechange) thatresults from this changingmix. 1955 1960 1965 1970 1976 Changeper point of weighted unemployment 0.82 0.82 0.87 0.95 1.02 Changeper point of conventional unemployment 0.82 0.82 0.85 0.87 0.88 In terms of a point of weightedunemployment, the responsivenessin- creasedby aboutone-quarter between 1960 and 1976. The changeis less markedin terms of conventionalunemployment rates; the two changes are differentbecause the cyclical relation between conventional and 9. Arthur M. Okun, "Upward Mobility in a High-Pressure Economy," BPEA, 1:1973, pp. 210-1 1. George L. Perry 19 weightedunemployment has also shifted over this period. The relation betweenthe conventionallymeasured unemployment gap (u - a) andthe weightedunemployment gap (u* - u) is estimatedby the equation

(2) u - u- = 0.996(u* - u7*) + 0.0127[T(u* - u*)], (65.7) (6.7) Standarderror 0.00072;Durbin-Watson = 1.9. whereT is a time-trendvariable starting in 1964 andthe numbersin parenthesesare I statistics(as they are in the equationsthroughout this paper). The levels of unemploymentused to define potential,u* and ui,are dis- cussedand defined below in connectionwith estimatingthe potentiallabor force. Equation2 is used to calculatethe second row of the table, which shows that the changingmix of the labor force accountsfor about one- quarterof the differencebetween the presentestimates of the cyclicaleffects on laborforce participationfor 1976 and those cited by Okun.The remainingdifference-between 0.6 percentand the 0.82 percentaverage for 1955-60 in the present estlinates-is attributableto the lagged unemploymenteffects and other differencesin the form of the estimating equations.

EXPLAINING THE TRENDS

Severalattempts were made to get behindthe time trendsshown in the table 1 equationsin the hope of improvingthe specificationsand of better identifyingthe cyclicaleffects and also of providinga strongerfoundation for projectingfuture participationrates. First, the proportionof black males in the adult age groupswas introducedas an explanatoryvariable, but did not accountfor the declinein these groups.Furthermore, participationrates for both black and white adultmales declinedin recentyears. Second, the proportionof 16- to 19-year-oldsenrolled in school was a significantvariable in explainingparticipation of femalesin this age group. However,it left both the time trend startingin 1967 and the unemployment effectsalmost identical to those in the table 1 equations,although it reducedthe time trendbefore 1970. The enrollmentvariable was insignifi- cantfor teenagemales. Finally, the ratio of college enrollmentsof women and men was tried as a proxy for the sociological changes affecting women's participation.But it resulted in substantiallyhigher standard errorsfor the equationthan the time trenddid and took on the wrongsign in somecases. 20 BrookingsPapers on Economic Activity, 1:1977 Table 2. Labor Force Participation Rates of Women, by Marital and Parental Status, and Age Groups, 1967 and 1975 Percent Marriedwomen, Otherever- Never- Parentalstatus husbandpresent marriedwomen marriedwomen and age group 1967 1975a 1967 1975a 1967 1975S Withno childrenunder 18 16-19 43.3 61.9 48.9 47.0 37.2 49.6 20-24 67.2 84.2 75.6 82.9 70.3 72.4 25-34 70.4 84.2 69.9 86.0 80.9 84.6 35-44 59.2 65.6 54.8 82.3 74.5 79.8 45 and over 31.8 34.6 29.7 29.3 47.4 49.3 Withchildren 6-17 16-19 ...... 20-24 45.2b 42.0 39.1 64.3 ...... 25-34 49.0 60.3 72.9 77.0 ... 35-44 46.1 57.5 68.55b 75.0o ... 45 and over 41.3 47.1 59.8 60.5 ... Withchildren under 6 16-19 20.9 34.9 27.6 44.2 20-24 29.3 40.8 52.2 64.5 ...... 25-34 25.5 39.0 52.5 59.2 ...... 35-44 26.9 35.5 51.8 47.1 ...... 45 and over 26.8 31.7 47.5 53.3 ... Source: U S. Bureau of Labor Statistics. a. 1975 data are adjusted for the difference between the aggr-egateunemployment rates for 1967 and 1975 usinig the equations developed in the text. Unlike the other data in this paper, the data in this table are not adjusted for the 1970 Census and are not adjusted to an establishment-employment basis. b. The 1966 value was used dtueto an apparent error in the 1967 figures. FamilyStatus. For women,family status has been an importantfactor determininglabor force participation.The importanceof changesin status for the recentchanges in participationtrends was examinedwith the data in table 2. There,participation rates are given for 1967 and 1975 for the femalepopulation disaggregated according to threecategories for marital status,three for numberof children,and five for age. To eliminatecyclical effectsin comparingthe two years, the raw data for 1975 were adjusted for the differencein weighted unemploymentbetween 1967 and 1975 usingthe equationsof table 1. Participationrates rose between 1967 and 1975 in almostall of the thirty-threecells but rose only a little for women 20 andover who hadnever been marriedand for women45 andover. The proportionof women in differentcells has shifted along with the changesin behaviorfor each cell. Table 3 showsthe relativeimportance of these two sourcesof total change: (1) the change that would have oc- George L. Perry 21 Table 3. Population-Shiftand Behavioral-ChangeEffects on Labor Force ParticipationRates of Women, by Age Group Percentagepoints Changein participationrates, 1967-75, by source Change Shift withinfamily- amongfamily- Age group Total characteristiccells" characteristiccells 16-19 12.9 12.7 0.1 20-24 13.6 8.9 4.7 25-34 16.2 11.9 4.8 35-44 12.1 10.2 2.2 45 and over 2.0 2.0 0.0 All groups 8.7 6.9 1.8 Source: Calculated from table 2. Figures are rounded. a. For explanation, see text. curredhad the populationdistribution among cells remainedunchanged whileparticipation rates in each cell changedas they actuallydid between 1967 and 1975 (in the table, this part of the total change is labeled "changewithin family-characteristic cells"); (2) the change that would have occurredhad participationrates in each cell remainedunchanged while the populationdistribution among cells changedas it actuallydid between 1967 and 1975 (in the table, this part of the total change is labeled"shift among family-characteristic cells") . Althoughthe effectof the secondkind of change-in the distributionof family status with participationrates fixed-is noticeable, particularly for the groupsbetween 20 and 34, the changesin behaviorwithin each cell betweenthe two yearsis easily the more importantsource of the observed shiftin participationrates. Of course,both sourcesof changecan reflecta commonsociological trend. A woman'sdecision not to marryor not to have childrencan have the samemotivations that lead her to join the work force. The resultsin tables2 and 3 permitno hardconclusions, but they invite a few conjectures.First, the far greaterimportance of changesin participation for a given grouprelative to populationshifts amonggroups suggests some permanencein recent developments.Even if the distributionof variouskinds of family status moves back towardits earlierpattem because marriageand child-bearingturn out to have been postponedrather than forgone,large changes in participationwould remain. Second, the unevendistribution of changesin participationsuggests that they arebased on sociologicalfactors rather than on shorter-runeconomic developments 22 Brookings Papers on Economic Activity, 1:1977 inflationand real earnings.All groups should have been affectedby economic developments,yet only women under 45 displayedsubstantial changesin behavior,either on balanceor withinfamily-characteristic cells. This findingis supportedby the time trendsin the equationsof table 1. Third,there is roomfor a substantialfurther increase in overallparticipa- tion ratesof women.The age groupsthat have been affectedwill continue in the working-agepopulation for many years, while those 45 years and over who have not been affected (or who show a decliningtrend in the resultsof table 1) will be leavingit. Women25 to 34 withoutchildren may have approachednormal limits in their participationsince the participation ratesof never-marriedwomen are matched by those for the two other cells. But apparentlythere is still scope for higher participationamong women with childrenif their behaviorcontinues, as it has over the past decade, to narrowthe gap between those with and without a husband present.

THE ROLE OF WAGES AND PRICES The model of labor force participationof equation 1 is obviously incomplete.One of its sins of omissionis its failureto take explicitaccount of wagesand prices. I triedto incorporatethem, drawing on an alternative model of participationdeveloped by Wachter.10The variableshe used werethe realwage, W, and the pricelevel, P, andmoving averages of their past values, W* and P*, which were used to representthe "permanent" wage and the "perceived"level of prices.Wachter uses the perceivedreal wageand the ratioof the currentto the perceivedreal wage as explanatory variablesto modelthe effectson laborsupply of permanentand transitory movementsin real wages. The ratio of the currentto the perceivedprice level measuresthe moneyillusion that fools workersinto enteringthe labor market.The expectedsign on the price-ratiovariable is positive,while the expected sign on the two wage variablesdepends on the group whose participationis beinganalyzed. In order to examine their effects on participation,the variablesthat Wachtersnecified were addedto those of equation1 .1 Table 4 gives the 10. Wachter,"Labor Supply Model." 11. W* is an eight-year moving average of W, the real wage, defined as total private nonagriculturaladjusted hourly earnings deflated by P, the consumer price index; and P* = O.4P -4- O.3P(-l) + O.2P(-2) + O.IP(-3). c0U r - o bio o G o

~~4)~~~~ ~ n O C-40 ,-% 5 5 I 4)4 E .a I Ii I I ?i

*. < CM 00 N. Xt 0 - If 4)L. e 0 0O 0 0 S:2z *t m o N O W N O

' If) OC)I~ I C) r - 0'0 | |'r0 CMN CD0 - ' 4 - ^M & ^

4) eS If) i M IM If) C) cj C-I i 'i O- O 1- . O .53 1 fl.I S I C O

C)) F- Zo~- ti X O I~~ O I I ; Do ,C: . . ? . o.o r:

ma So v)- M -o -0 - h zC

OO 0 O~~~~~~~~%|0 | tn X > Q ~~~~~~~~N WPWtO?Cd N C8 2 m r .0 z4 < EOIgI * 4) 4) I . * a Di )

o 40 -l _ _|_|_| 0 M 4> |1C ~ ~ ?Q Q14 *|) tn en in | M Cm |M CD 00 t0

' ~~~q 0~~~~~~W0 4)

~~~f)~~~~~~CM . ~ ~ ~ .

0' K~~~~~~~~~~~~~~~~~~~~K~~~

4) '- %-) k0n4 ~4

I I CM~~~ '-~ i I101 4) ii I4-d14--.14-14-.k 1t 4 O 4 w-4 V-1m 1* W 24 BrookingsPapers on Economic Activity, 1:1977 estimatesresulting from fitting this equationto annualdata over the period 1954-76 for thosedemographic groups that Wachter used in his own tests. The yearsbefore 1954 wereunavailable because of the long lags required in formingthe variables.But the inflation since his stopping point of 1968 should provide a good test of the role of wages and prices in the modelfor laborsupply. The resultsin table 4 supportthe basic model of equation 1 in most respects.Each of the new variableswas includedif its t statisticwas at least 1.0. Despitethis lenientcriterion, most of the variablesreflecting the relativereal wage were omitted as insignificant.Only two of six actual real-wage variables were significantand correctly signed, while four othershad significantcoefficients with incorrectsigns.'2 The inflationvari- ableswere much more successful, with significant coefficients appearing in five of the eightequations. As the last two columnsof table 4 show, the additionof the wage and price variableshad only a modest effect on the cyclical responseof the model. They do producenoticeable changes in the time trendsfor some groups. In particular,compared with the results of table 1, the trend growthin participationis noticeablyfaster for females between 20 and 44 andfor maleteenagers.

CHANGING PERMANENCE OF PARTICIPATION

One final hypothesisconcerning cyclical responsewas checkedout. If the rise in participationrates of the past decadewas partof a moreperma- nent attachmentto work on the part of secondaryworkers, the cyclical responsivenessof participationcould be expectedto decline. This would meanthat the averagecoefficients on unemploymentshown in table 1 were too largefor the labor force of recentyears. But table 5, which gives the unemploymentcoefficients of the table 1 equations separatelyfor the 1954-65 and 1966-76 periods, suggests that this is not the case. The coefficientsare actuallylarger (more negative) for six of the ten population groupsin thelatter period.

12. Since the signs on the wage and relative-wage variables vary according to which theory applies, I have shown all significantcoefficients and have used Wachter's criteriaas expressedin his results and discussionto determinewhat signs are expected. George L. Perry 25 Table 5. Comparisonof Cyclical Responsivenessof Labor Force Participation, by Age-Sex Group, 1954-65 and 1966-76 Sum of unemploymentcoefficients Sex and age group 1954-65 1966-76 Males 16-19 -2.69 -3.04 20-24 -0.80 -1.47 25-34 0.03 -0.12 65 and over -2.38 -3.03 Females 16-19 -2.95 -3.09 20-24 -3.10 -2.37 25-34 -2.23 -1.67 35-44 -1.79 -1.29 45-54 -0.23 -0.52 55-64 -1.00 -0.52 Source: Equations of table I were used, with estimated effects of right-hand variable other than unemployment subtracted from the dependent variable and the remaining unemployment effect estimated sepa- rately for each period shown.

PotentialLabor Input

Since their unemploymenteffects seemed reasonablystable through time and were much the same whetheror not price and wage variables were included,the participationequations of table 1 were used to adjust for the cyclical variationsin the size and compositionof the labor force and employment.For this purpose, a constantweighted unemployment rate of 3.32 percent-its value in mid-1955-was taken as an index of potential, with cyclical variationsmeasured by deviations of weighted unemploymentfrom this constantlevel.

DEFINING POTENTIAL Since a definitionof potentialis needed and is being introducedhere, this is as good a place as any to explainthe choice. Because everyoneis used to it and understandsit, I departonly reluctantlyfrom a constant conventionalunemployment rate as the index of the economyat potential. Yet the weightedunemployment rate has the obvious advantagethat it 26 BJ3ookingsPapers on Economic Activity, 1:1977 representsunemployed labor in efficiency units. And its counterpart, weightedemployment, has proven useful in aggregateproduction func- tions. One might want to define potentialby a measurethat representeda "noninflationary"level of utilizationor unemployment.An appropriate level of constantweighted unemployment is a step in that direction.But it wouldbe misleadingto thinkof such a measureas defininga noninflationary operatingpoint with any precision,and it is not meant to serve that purpose. Even when I first used it, I noted that weighted unemployment was an imperfectmeasure of overall labor market tightnessbecause it failedto accountfor the changingdispersion of unemploymentrates. And everyonenow knowsthat inflation can arisefrom causesother than overly tight labor markets,and that once under way, it can continue despite enormousslack in capacity and high unemployment.Furthermore, if policies directlyaffect the price-wagelevel-through structuralmeasures aimedat high unemploymentareas, incomes policies, or microeconomic policies-the link betweenweighted unemployment and inflationis weak- enedfurther. Analysis of potential output provides informationabout the relation between output and unemploymentand about the cyclical and trend characteristicsof productivity,labor input, and outputin the economy.An understandingof these relationshipsis importantto any macroeconomic investigation.The concept of weightedunemployment that is used in this analysisapproximates overall tightnessin labor marketsin a way that is usefulfor studyinginflation and certainlymore useful than conventional unemployment.But using a level of constantweighted unemployment as a base for measuringpotential does not imply that that base gives the optimaloperating level oI the economy at all times, or that inflationwill always be absent or declining if that rate is maintained.Choosing the optimallevel thatbalances risks of inflationagainst costs of unemployment and underutilizationwill continueto be an issue for year-by-yearanalysis andfor debatewithin the politicalprocess.

POTENTIAL EMPLOYMENT AND HOURS The potentiallabor force in each age-sexgroup (Li) in each year was calculatedby addingto the actuallabor force the cyclicalincrement pre- George L. Perry 27 dicted by the table 1 equations.The unemploymentrate at potentialfor eachgroup was then estimated using the restrictionthat Us+ Li;- Li Ul Li n- Li for each group,where U is unemploymentand bars over the variables representvalues at potential;that is to say, the ratio of the actual "total" unemploymentrate (which includesthe shortfallin participants)to the unemploymentrate at potentialis the same for each group.The constant a was calculatedfor each year from the constraintthat the weightedun- employmentrate was 0.0332:

a?3gihiw 'Lii = 0.0332, wherethe fl arethe relativelabor force weightsof 1966; the hi and wi are the hour and wage weights,respectively; and E/3ih,wi = 1. Employment at potential for each group is Es = i - L; total employment is E = YEl; and total weighted employment is WE = Yhjw,Ej. Along the potentialpath, the unemploymentrate has risen gradually over the past twentyyears, as shown in table 6. There, the armedforces are excludedto make the data comparablewith the faamiliarpublished unemploymentrates.13 The unemploymentrate at potentialis estimatedas 4.9 percentin 1976, the same as CEA's estimate, comparedwith 4.0 percentin the mid-1950s. Unemploymentwas clearlybelow its potential level in two sustainedwartime stretches, from 1951 through 1953 and from 1966 through 1969. It was about at potential in 1955-57 and in 1965 and 1973. In the fifteenremaining years, unemployment was above potentiallevels, withthe biggestexcess occurringin the last two years. EmploymentGaps. Table 7 shows the percentemployment gap estimatedfor eachdemographic group per point of aggregateunemployment. The relativegains in employmentexperienced by each group are slightly differentin 1960-61 and 1975-76 becauseof differencesin theirrelative sharesof total employmentand unemploymentin the two periods.But in

13. In this table and all subsequent tables with economic variables on the potential path, potential values are given by three-year centered moving averages of the directly calculated potential variables. This removes some of the irregularitiesthat would otherwise arise from residuals in the estimated equations that are used to calculate potential magnitudes, and from abrupt changes introduced by time trends. 28 Brookings Papers on Economic Activity, 1:1977 Table 6. ConventionalUnemployment Rates, Actual and Potential, 1950-76 Percent Actual minus Year Actual Potential potentiala 1950 5.2 4.0 1.2 1951 3.1 3.9 -0.8 1952 2.8 3.7 -0.9 1953 2.7 3.6 -0.9 1954 5.3 3.8 1.6 1955 4.2 4.0 0.3 1956 4.0 4.0 0.0 1957 4.1 4.0 0.2 1958 6.7 4.0 2.7 1959 5.4 4.1 1.3 1960 5.4 4.1 1.3 1961 6.6 4.1 2.5 1962 5.5 4.2 1.2 1963 5.6 4.3 1.3 1964 5.1 4.4 0.8 1965 4.4 4.4 0.0 1966 3.6 4.4 -0.8 1967 3.8 4.4 -0.6 1968 3.6 4.5 -0.9 1969 3.5 4.5 -1.0 1970 5.0 4.5 0.4 1971 6.0 4.6 1.3 1972 5.6 4.8 0.9 1973 4.8 4.8 0.0 1974 5.6 4.8 0.8 1975 8.5 4.8 3.7 1976 7.7 4.9 2.8 Sources: Actual from Bureau of Labor Statistics, with rates adjusted for Census-year changes and 1967 definition changes. The calculation of unemployment rates at potential is described in the text. a. Calculated from unrounded unemployment rates.

bothperiods the storyis muchlike Okun'scalculations based on my earlier analysisof potential.'4 EmploymentComposition. Throughtime, the changingdemographic compositionof employmenthas been the resultof two developments:the changingcomposition of the labor force and changingrelative unemployment rates. Since, for the most part, those groups that have grown the 14. Okun, "UpwardMobility," pp. 216-17. George L. Perry 29 Table 7. EmploymentGaps by Age-Sex Group per Point of the UnemploymentRate, 1960-61 and 1975-76 Percent Sex and age groups 1960-61 1975-76 Males 16-19 5.4 5.5 20-24 2.6 3.0 25-34 1.2 1.2 35-44 0.9 0.8 45-54 1.1 0.8 55-64 1.3 0.8 65 and over 2.5 2.3 Females 16-19 5.3 5.9 20-24 3.7 3.7 25-34 2.8 2.6 35-44 2.3 2.0 45-54 1.6 1.3 55-64 1.5 1.4 65 and over 0.7 0.9 Source: Author's estimates based on participation equations and calculation of potential employment described in the text. fastesthave also experienceda deteriorationin unemploymentrates relative to othergroups, the compositionof employmenthas changedslightly less than the compositionof the labor force. The employmentalong the economy'spotential path differsfrom the actual to the extent that the cyclical employmentgains shown in table 7 are not proportionalamong the variousdemographic groups. Tab'le 8 gives employmentprofiles of the economy,showing total employmentand its demographicdistribution for the actualeconomy and the estimatedeconomy at potential.The estimate for 1981 in the tablewas obtainedby projectingparticipation rates beyond 1976, makinga conservativeprojection of the participation-rateequations in table 1. Ratherthan projectinga continuationof the estimatedtrends, whichwould maintaina constantpercentage change, the averageannual absolutechange in participationrates over the past ten yearswas extended into the future. The resultingparticipation rates for each demographic groupwere multiplied by 1981 populationsto obtain 1981 projectionsof the laborforce. These, in turn,were multiplied by 1976 potentialemploy- mentrates to obtain1981 employmenttotals for each group. The work force is maturing.The large increasesin employmentpro- 30 Brookings Papers on Economic Activity, 1:1977 Table 8. Actual and Potential EmploymentProfiles of the Economy, Selected Years 1955-81 Projected Description 1955 1960 1965 1970 1975 1976 1981 Actualeconomy Unemployment rate (percent), 4.2 5.4 4.4 5.0 8.5 7.7 ... Total employment (millions)a 65.8 67.9 73.4 81.9 86.4 89.0 Employment propor- tion (percent) 65 and over 5.0 4.7 4.0 3.8 3.2 3.0 ... 16-19 6.4 6.6 7.5 8.0 8.5 8.5 ... 20-24 10.7 10.3 11.9 14.0 14.3 14.5 25-64 Female 23.0 24.3 24.8 26.1 27.5 28.0 ... Male 54.8 54.1 51.8 48.0 46.5 46.0 Potentialeconomy Unemployment rate (percent)a 4.0 4.2 4.4 4.5 4.9 4.9 5.1 Total employment (millions)a 66.3 69.7 74.0 82.2 92.0 94.6 106.2 Employment propor- tion (percent) 65 and over 5.1 4.7 4.0 3.8 3.1 3.0 2.6 16-19 6.5 6.9 7.5 8.1 9.4 9.4 8.9 20-24 10.8 10.5 11.9 14.1 14.9 15.1 15.4 25-64 Female 23.0 24.4 24.8 26.1 27.4 27.9 30.0 Male 54.7 53.6 51.7 48.0 45.1 44.6 43.1 Source: Actual-Bureau of Labor Statistics; other-calculations underlying table 7. Figures are rounded. a. Unemployment rates exclude armed forces; employment totals and proportions include them. portionsfor the groupsaged 16 to 24 thathave been occurringsince 1960 are over exceptfor the normalexpansion in theirshare of jobs as cyclical unemploymentis reduced (unless special governmentjob programsadd to their expectedjob total). The rise in the employmentshare of women aged25 to 64 will actuallyaccelerate as morewomen enter this matureage groupand participationrates continuerising although at a more modest rate than in recent years. Finally, the share of men 25 to 64 in total employmentdeclines more slowly than in recentyears alongthe potential path, althoughthe decline is substantialbetween 1976 actual and 1981 potential.These relative changes suggest that the specific problems of George L. Perry 31 teenagersand youngadults may lessen, while adultwomen may be in for a relativelymore difficult period in the labormarket as theirnumbers continueto increasesharply. Potential Average Hours. Average hours worked per year per em- ployeefluctuate cyclically and, in addition,have been on a cleardownward trendthroughout the postwarperiod. Both the cyclicaland secular changes reflect the operationof several forces. Up throughthe mid-1950s, the normalworkweek in manufacturingwas beingreduced to 40 hours.Since then, a gradualreduction in economy-wideaverage hours has continued, onlyin partas the resultof the growingnumber of workerswho voluntarily seekpart-time work. My previousattempt at explainingthe trendin hours adjustedfor the changingcomposition of employmentstill left a major portionunexplained.'5 So this timeI have ignoredthat complication. Since the subsequentanalysis will concentrateon the private nonresidential nonfarmbusiness sector-hereafter, simply the "businesssector"-for the period startingin 1954, the equationwas fitted to averagehours in thatsector (AHB) from 1954 to 1976:

(3) log AHB= 3.72 - 0.494u*- 0.00214T54-0.0031OT66- 0.00885DT. (-5.5) (-8.1) (-5.6) (-4.3) Standarderror = 0.00324; Durbin-Watson= 2.1. Threetime-trend variables were needed: T54 startsat 1 in 1954 and rises by 1 eachyear; T66 startsat 1 in 1966 andrises by 1 each year;and DT is 1 in 1966 and 2 in 1967 and subsequentyears. Using a special time dummyfor 1966 and 1967 is unsatisfying,but the data demandit. I have previouslydiscussed my attemptto get behindthis discontinuity,concluding then that the drop in hours should be regardedas a permanentshift downwardin the hourstrend.'6 Now, with seven moreyears of data available, that conclusionseems reaffirmed.Nonetheless, estimating the cyclical and secularbehavior of hourswith an equationsuch as 3 remainsone of the leastsatisfactory steps in the analysisof potential. The coefficienton u* indicatesthat each extrapoint of weightedunem- ploymentis accompaniedby nearly 0.5 percent lower averagehours in the businesssector. In 1976, that correspondsapproximately to 0.4 percent extra averagehours per point of conventionalunemployment. The trendterms indicate that, at present,average hours are fallingby just over

15. Perry, "LaborForce Structure." 16. Ibid. 32 Brookings Papers on Economic Activity, 1:1977 0.5 percenta year along the potentialpath. With the changesin relative employmentamong demographic groups just projected, this rateof decline couldlessen, but the equationprovides no way of estimatingthat effect. Potential WeightedHours. The ingredientsfor calculatingpotential weightedhours are now at hand. Employmentin each age-sex group was allocatedto the businesssector accordingto the ratio in each year of total employmentin that sector to economy-wideemployment. Employ- mentin the businesssector in each age-sexgroup, EBi, was then weighted by relativehours and wages and summedto get weightedemployment for the sector, WEB, and adjustedby averageweekly hours to get weighted (total) hours, H, the basic measureof labor input.'7The differencein employmentbetween actual and potentialwas all assignedto the business sectorand the procedurerepeated to obtainpotential weighted hours, H, for eachyear.'8 The averageannual growth rates in principalindicators of potential laborinput are shownin table 9 over half-decadeintervals and 1970-76. The last columnprojects the next five years.The growthrate of the labor force and employmentmore than doublesin the 1965-76 periodfrom the previousdecade, and growth in potential employmentaccelerates even more in the business sector than it does in the whole economy. The growthrate of weightedhours in the business sector does not speed up nearlyas much as employment,both because of the sharp slowdownin averagehours worked during that period and because a large part of the employmentacceleration consists of workerswith low hours and wages. But with averagehours decliningmore slowly in the 1970-76 inteival, potentialweighted hours grow more than twice as fast as in the 1965-70 period. Because the potential employment increase continues to be relativelyconcentrated in workerswith low weights (and averagehours declinefaster than in the pre-1965 intervals),the annualgrowth of potential weightedhours is a full percentslower than the growthof potential employmentover the latest interval,a much largerdifference than in the pre-1965periods. For the 1976-81 projectionperiod, total employmentand its composi- 17. The following formula was used to assign the appropriateweights consistent with the total hours of employment: H = AHB hEBi hiwiEBi. MhiEBi 18. A justificationfor treating the rest of the economy as noncyclical is provided by Okun'sresults in "UpwardMobility," pp. 218-24. George L. Perry 33 Table 9. Potential Growth Rates of Labor Input Measures, Selected Intervals,1950-81 Percentper year Projected Measure 1950-55 1955-60 1960-65 1965-70 1970-76 1976-81 Labor force 1.43 1.01 1.29 2.17 2.39 2.08 Employment Total 1.45 0.97 1.23 2.14 2.31 2.07 Total civilian 1.05 1.17 1.22 2.10 2.59 ... Business 1.46 1.21 1.11 2.30 2.87 2.07 Hours Total business 1.15 0.95 0.84 1.41 2.33 1.53 Weightedtotal business 1.06 0.81 0.59 0.94 1.81 1.39 Source: Calculated as described in the text. tion aretaken from table 8 while averagehours are projected using the rate of declinegiven by equation3. The ratio of businessemployment to total employmentwas maintainedat its potentiallevel in 1976. This assumption is probablya conservativeprojection of the portionof employmentin the high-productivitybusiness sector since the ratio of businessemployment to eithertotal or total civilianemployment generally rose in the past.

Productivity

The relationshipbetween labor input and outputin the businesssector providesthe basisfor examiningrecent productivity behavior and estimating a potentialoutput path for the econiomy.I startwith the model relating output and weightedlabor hours that I used previouslyfor the whole economy'9and apply it here to the businesssector.20 I shall turn to the issueof capitaladequacy below. The basic model starts out with the propositionthat weighted labor productivitygrows exponentially along the potentialpath:

(4) Q = Bert

19. Perry, "LaborForce Structure." 20. Nonresidential nonfarm business product equals total business product less product originating in farming and services of the housing stock. The scope is the same as in Edward F. Denison, Accounting for U.S. Economic Growth (Brookings Institution, 1974), and corresponds as closely as possible to the concept for labor input that is used in this paper. 34 Brookings Papers on Economic Activity, 1:1977 where Q is output in the business sector, r is the annual growth rate, t is a time index, and the bars over variablesindicate potential values. This can be modifiedto allowfor a breakin the growthtrend:

(S) Q FI = Be(r tl+r2t2). Cyclicaldeviations of productivityfrom its trendare expressed by (6) Q= (Hj Q H where/3 > 1 if, as expected,productivity is higherthe higherthe level of actual hours or actual output relative to potential. Previous work has shownthat some lags exist in this cyclicalrelation, and they are allowed for by modifying6 to

(7a) H_ _ Q/Q-p1 H Q2 L(QIQ)_12 or to (7b) Q _ (ftryH/HJ1 Q VHJ (/)1 Combiningequations 5 and either 7a or 7b to eliminateQ leads to the modelused for statisticalestimation.

RECENT BEHAVIOR OF PRODUCTIVITY In 1974, at the start of the big recession,actual labor productivityin the business sector (in terms of conventionalhours, not the weighted hoursthat areused throughoutmost of this analysis) fell 3.6 percent.This was not only a huge decline,but also the firstone of any size in productivity sincethe war.Partly making up for this, productivityrose in the following year a substantial1.8 percentdespite the deepeningrecession. How much should the behaviorof productivityin 1974 influenceany assess- mentof the longer-runcharacteristics of the economy? There is ample evidence that businessmenmisjudged the severity of the downturnfor a long time. Theykept expandingemployment long after outputfell. And they continuedto build up inventorieslong after final sales collapsed.Their optimism may havebeen encouragedby the government and the press,both of which continuedto foster the impressionthat the economywas experiencingan "energyspasm" rather than a truereces- George L. Perry 35 Table10. TwoEquations Relating Output and Labor Input, withand without Trend Break No trendbreak Trendbreak in 1969 Coefficient, regressionstatistic, and Equation Equation Equation Equation structuralcharacteristic JO.P 10.2b JQ.ja 10J. Coefficient ao -4.938 -8.102 -5.279 -8.113 a, 0.0163 0.0267 0.0179 0.0279 (20.5) (72.4) (13.5) (43.3) a2 0.609 1.573 0.651 1.438 (20.9) (17.6) (16.5) (14.0) a3 -0.087 0.149 -0.093 0.150 (-2.1) (1.5) (-2.3) (1.7) a4 -0.0262 -0.0514 -0.0237 -0.0421 (-3.6) (-4.3) (-3.3) (-3.6) at, ...... -0.0019 -0.0042 (-1.5) (-1.6) Regressionstatistic Standarderror 0.0064 0.0112 0.0062 0.0102 Durbin-Watson 1.82 1.64 1.77 1.55 Structuralcharacteristic (percent) (dQ/Q)/(dH/IH) Permanent 1.64 1.57 1.54 1.44 First year 1.92 1.72 1.79 1.59 Weightedproductivity trend (percent) Before 1969 2.68 2.67 2.75 2.79 Since 1969 2.68 2.67 2.46 2.37 Weightedproductivity residual (actual minus predicted in percentagepoints) 1976 -0.6 -1.1 -0.3 -0.4 1975 -0.4 -1.2 -0.3 -0.9 1973 -0.1 -0.2 -0.2 -0.6 Source: Estimated from equations 10. 1 and 10.2 given in notes a and b below. The period of estimation is 1954-76. a. Equation 10.1:

log (H) ao + alT54 + a2 log (T) + as [log (g) -log (H)g] + a4D74 + a5T69,

where H is weighted total hours, H is potential weighted hours, Q is business-sector output, T54 and T69 are time-trend variables, and D74 is a dummy variable whereby 1974 is ignored. b. Equation 10.2:

log ( Q) = ao + a1T54 + a2 log ()_ + as [log (_) - log (4_) ] + a4D74 + a5T69.

sion. But regardlessof whetherthese assessmentswere reasonableat the time,viewed in retrospect,they led to extremelyunusual cyclical behavior 36 Brookings Papers on Economic Activity, 1:1977 in the economy. It seems best to acknowledge this, ignore 1974, and in- vestigate recent productivity with the benefit of 1975 and 1976 developments. In the estimates of the model shown in table 10, this is accom- plished by introducing a dummy variable, D74, for 1974. The two estimating equations reported in table 10 use, alternatively, weighted hours (equation 10.1 from 5 and 7a) and output (equation 10.2 from 5 and 7b) as dependent variables. The relation between the equation coefficients and the structuralparameters of the miodelare as follows:

Equation 10.1 Equation10.2 Weightedproductivity growthtrend Withouttrend break a1/a2 a, With trendbreak (a, + a5)/a2 a, + a4 Elasticityof output gap to inputgap Firstyear 1/(a. + a3) a2 + a, Permanent 1/a2 a2 Constant Productivity Trend. To start with, consider the equations estimated with no break in the time trend over the 1954-76 period. The constant annual trend in weighted productivity is 2.68 percent and 2.67 percent in equations 10.1 and 10.2, respectively. Equation 10.1 predicts a permanent increase of 1.64 percent in output for each percent of gap- closing increase in weighted hours, or a weighted productivity bonus of 0.64 percent for each extra percent of weighted labor input. In the first year, the productivity bonus is 0.92 percent. The productivity residuals from the equations that are displayed in the last three rows of table 10 all show that productivity was overestimated in recent years by the historic equations, but not by much. To put these errors in perspective, in table 11 the recent overpredictions from the equations that were estimated without a break in the productivity trend are compared with residuals from earlier years. The table gives the residual for all years in which it exceeded the root mean-square error (RMSE) of the equation for both forms. The residuals are shown as percentage shortfalls in actual weighted productivity from the predictions of the equation. The 1974 shortfall is huge in both forms of the equation. (It is not an estimated residual but the shortfall indicated by the dummy variable for that year.) The shortfalls in subsequent years are substantial compared with the RMSE of the equations, but so were the errors in seven of the George L. Perry 37 Table 11. Large Residualsfrom ProductivityEquationsa Percent Year Equation 10.1 Equation 10.2 1956 -1.0 -1.8 1957 -1.2 -1.9 1958 1.1 1.4 1960 -0.6 -1.3 1969 -0.7 -1.2 1971 0.6 1.1 1972 0.8 1.5

1974 - 2. 6b _ 5. lb 1975 0 -1.2 1976 -0.6 -1.1 Source: Equations with unbroken trend from table 10. a. All residuals with absolute values larger than the root mean-square error of the residuals from the equations are shown. b. 1974 was dummied out in the equations. The errors shown here are the errors implicit in the values of the dummy variable. c. Under the stipulation of note a, the 1975 residual (0.4) was too small to be included. twentyyears before 1974. By this standard,the presentperiod is unusual, butnot unusuallyunusual. The recentperformance of productivitymay appearweaker than these shortfallsindicate because of the substantialpositive residuals in 1971-72. Between 1972 and 1976, productivityadjusted for cyclical effects rose 2.5 percentless than expectedfrom equation 10.2 and 1.4 percent less usingequation 10.1. Even these swingsare not extraordinarywhen comparedwith the swingsbetween 1957 and 1958, 1958 and 1960, or 1969 and 1971. Of course, in those cases, we know that subsequentyears broughtthe errorsback to zero;we do not knowthat yet thistime. All in all, the equationsestimated with a constantproductivity trend can hardlybe rejectedon the basis of the economy'srecent performance: 1974 was an unusualyear that requiresad hoc explanations.But most of the productivityweakness that remainedby 1976 is accountedfor as the normalshortfall of productivitythat accompanieshigh unemployment. Thisis especiallytrue using equation 10.1. On the otherhand, the constant productivitytrend shown in table 10 is slower than the trend that would havebeen estimatedwith equationsending several years earlier.The pos- sibilityof a slowdownin the trenddeserves further exploration. Slowdown in Productivity Trend. The way to maximize one's concern about recent developmentsis to isolate the latest years. Equation 10.2 38 Brookings Papers on Economic Activity, 1:1977 estimatedthrough 1973 showsa 3.1 percentshortfall in productivitywhen projectedto 1976. But this procedureseems biasedtoward discovering a permanentproductivity slowdown. It seemedbetter to explorethe hypo- thesisof a trendbreak by allowingfor one after a fairlyrecent year with a smallprediction error. I chose 1969 to starta new trendand the resulting estimatesare shownin the righthalf of table 10. The equation10.1 form showsa 0.3 point slowdownafter 1968 in the productivitytrend, and the equation10.2 form a 0.4 point slowdown. Comparedto the equations without a trend break, these equations show faster productivitytrends before 1969 and smallercyclical productivity effects. The new equations have slightlylower standarderrors than those withouta trendbreak and show somewhatsmaller prediction errors for 1975 and 1976. Equation 10.1 estimatedwith the broken trend shows not only small but stable residualsin theseyears and in 1973, suggestingthat its trendestimate over thatinterval is close to the mark. PossibleSources of a TrendBreak. In view of the mixed evidenceon recentproductivity trends provided by the basic model, I looked for some alterationin the specificationthat would help resolvethe issue. Like most other investigators,I suspect that a more fully specified model would include a role for the capital stock; but unlike some others, I did not choose to force the capitalstock into the storyby constrainingit to enter the equations.I tried to find an explicitrole for it in this relationshipby adding,in turn,several versions of it to the basic model and estimatingits effect.Expenditures for pollution abatementwere subtractedfrom gross nonresidentialstocks in the nonfarm business sector. Structuresand equipment,and equipmentseparately, each for the total sector and for manufacturingseparately, were tried as capital-stockmeasures, but each failedto enterthe relationshipsignificantly. Anotherpossible source of a slowdownin the productivitytrend as I measureit wouldbe that the weightingused for young workersoverstates theircontribution to output,perhaps because firms are investing in training them. Or more generally,the wage weights may simply be inaccurate proxies for relative productivities.However, neither the proportionof teenagersnor the proportionof adultmales was successfulas an additional explanatoryvariable. Since I could not settle to my own satisfactionthe questionof whethera trend break was appropriate,I used results both with and withoutthe break to form two separateestimates of potential businessoutput and potential GNP. George L. Perry 39

PotentialGNP

Potentialbusiness output, Q, was generatedby applyingthe trendpro- ductivityestimates from each form of equation 10.1 to the estimatesof potentialweighted manhours and anchoringpotential output by equating it to actualoutput in the middletwo quartersof 1955; this is the bench- markfirst established by Okunand widelyaccepted since. PotentialGNP was then calculatedby addingthe differencebetween actual GNP and businessoutput in each year. This procedureallows no change between actual and potentialin either labor input or productivityfor the non- businesspart of GNP. For projectingpotential GNP, the same two productivitytrends for the businesssector were extendedand combinedwith the projectionof weightedmanhours shown in table 9. As noted earlier,these projections assumedthat the businesssector would maintainits 1976 share of total potentialemployment. The remainingemployment was assigned to the nonbusinessportion of GNP and the output that would be forthcoming therewas projected using a relationestimated from the 1954-76 period:

(8) log QN = 0.163 + 0.488 log EN + 0.0256TS4, (2.0) (8.7) Standard error = 0.0152; Durbin-Watson = 1.678; p = 0.60. whereQN andEN are outputand employmentoutside the businesssector and T54 is a time-trenddummy. The variablesQN and EN cover the governmentsector, in whichthere is no productivitygrowth by definition; the servicesof housing,which are producedwithout any labor; and agri- culture,where productivity growth is very rapid.This is such a mixedbag thatI acceptedthis equationwithout question. The two estimatesof potential GNP are shown in table 12, together with the CEA's revised estimates and the previous official estimates. PotentialI, which is based on the slower productivitytrend startingin 1969, growsat a 3.9 percentannual rate between1976 and the 1981 pro- jection;potential II, basedon the constantproductivity trend, grows about 0.2 point a year faster. In this projectionperiod, both measuresgrow at aboutthe same rate as they did in the 1970-76 interval.In earlieryears, potentialI grew slightlyfaster than 31/2 percent a year and potentialII slightlyslower. 40 Brookings Papers on Economic Activity, 1:1977 Table 12. Potential Gross National Product, 1954-76 and Projected, 1977-81 Billions of 1972 dollars PotentialI PotentialII Officialestimates of (produictivity(unbrokent potential GNP gap trenidbreak productivity Year in 1969) trenzd) Revised Previous PotentialI PotentialII 1954 633.0 633.3 629.7 634.4 -19.3 -19.6 1955 654.8 654.8 651.4 656.6 0.0 0.0 1956 674.3 673.9 673.9 679.6 -5.5 -5.1 1957 697.6 696.9 697.2 703.4 -16.7 -16.0 1958 722.2 721.0 721.3 728.0 -42.6 -41.5 1959 749.5 748.0 746.2 753.5 -29.1 -27.6 1960 777.2 775.1 771.9 779.9 -40.4 - 38.3 1961 804.7 802.2 798.6 807.1 -49.4 -46.9 1962 832.9 829.9 826.4 835.4 -33.8 -30.8 1963 859.8 856.3 857.1 865.9 -29.1 -25.6 1964 890.0 885.7 890.3 898.4 -15.5 -11.3 1965 922.7 918.0 925.0 932.1 3.2 7.9 1966 956.0 950.7 960.8 967.0 25.0 30.3 1967 988.8 982.7 996.3 1,003.3 18.9 25.0 1968 1,021.9 1,015.8 1,031.7 1,040.9 30.0 36.0 1969 1,055.9 1,050.6 1,068.3 1,081.6 23.0 28.2 1970 1,095.4 1,091.7 1,106.2 1,124.9 -20.1 -16.4 1971 1,138.1 1,136.0 1,145.5 1,169.9 -30.6 -28.5 1972 1,185.1 1,184.8 1,186.1 1,216.7 -14.0 -13.7 1973 1,231.3 1,233.1 1,228.2 1,265.4 3.7 1.9 1974 1,279.6 1,283.6 1,271.7 1,315.9 -65.6 -69.6 1975 1,328.5 1,334.9 1,316.9 1,368.6 -136.7 -143.2 1976 1,379.2 1,388.1 1,363.6 1,421.2 -114.6 -123.5 1977 1,436.7 1,448.5 1,412.0 ...... 1978 1,492.3 1,507.0 1,462.1 ...... 1979 1,550.4 1,568.3 1,513.9 ...... 1980 1,607.9 1,629.3 1,567.7 ... 1981 1,668.3 1,693.3 ...... Sources: Potentials I and II are derived from table 10 equations and text equation 8; official estinmates are those of the Council of Economic Advisers from Economic Report of the President,Janiuary 1977, p. 54, except for 1977-81, which are from Peter K. Clark, "A New Estimate of Potential GNP" (Council of Economic Advisers, January 27, 1977; processed); GNP gap is calculated by subtracting potential I or II, in the first two columns, from the corresponding actual GNP from Economic Report, 1977, p. 54.

OKUN 'S LAW How well do Okun'slaw and the new estimatesof potentialGNP stack up againstone another?Table 13 showsseveral equations relating gaps in output and in conventionalunemployment rates. The latter are not the George L. Perry 41 Table 13. Okun's Law Estimates Using Potential Gross National Product I and Ia

Regressionstatistic Okun's Coefficient -Ou' Equationand Standard Durbin- law co- potential a1 a2 a3 error Watson efficient

Equation 13. 1b Potential I 2.88 0.16 0.031 0.0072 1.4 3.04 (21.0) (1.1) (4.3) Potential II 2.94 0.14 0.034 0.0085 1.1 3.08 (18.3) (0.8) (4.0) Equation13.2c Potential I 0.328 -0.010 ... 0.00238 1.6 3.05 (28.3) (-4.2) ... Potential II 0.319 -0.010 ... 0.00265 1.5 3.13 (25.5) (-3.8) Source: Based on data in tables 6 and 12. a. The numbers in parentheses are t statistics.

b. Equation 13.1: Q Q = ai(u- i) + a2(U-u)-1 + a3D74, Q where Q is real GNP and u the conventional unemployment rate; bars over variables indicate potential values; and D74 is a dummy variable for the year 1974. All equations were estimated with annual data over 1954-76.

c. Equation 13.2: (u-u) = al (Q Q) + a2D74.

differencebetween actual unemploymentand 4.0 percent, but those unemploymentgaps measuredagainst the changingunemployment rate thatdefines potential as shownin table 6. The equationsuse the outputgap and unemploymentgap alternativelyas dependentvariables. When the lagged unemploymentgaps are used as an added independentvariable, their coefficientsare only about equal to their standarderrors. The year 1974 is dummiedout on the grounds described earlier. The question usuallyasked of thisrelation is how muchunemployment will changefor a givenchange in the GNP gap. In the presentestimates, the answeris much the same no matterthe form of the equationin the table, althoughthe equationsin the bottomhalf of the table are moreappropriate in principle. Comparedwith Okun'soriginal estimate that a 3.2 percentoutput gap was associatedwith one point of unemployment,the presentestimates of the Okun'slaw coefficientare 3.05 using potentialI and 3.13 using potential II. Both the intervals1969-76 and 1973-76 are trackedreasonably well witheither potential estimate and its correspondingOkun's law coefficient. 42 BrookingsPapers on Economic Activity, 1:1977 In the longer interval,real GNP rose 17.2 percentcompared with 30.6 percentfor potentialI and 32.1 percentfor potentialII, thus openingup outputgaps of 11.2 and 12.6 percent,respectively. Okun's law predictsa wideningof 3.7 points in the unemploymentrate gap using I and one of 3.9 points using II; the actual rise was 3.7 points. Between 1973 and 1976, the predictedrise in the gap was 3.0 and 3.1 points, respectively, for potentialsI and II; the actualrise was 2.8 points. Startingfrom a year like 1971, with the large productivityresidual noted earlier,the relation does not performas well. The unemploymentgap widens by 1.5 points comparedwith predictionsof 2.1 points from potentialI and 2.3 points from potentialII. I interpretthese results as generally supportingthe potential estimates made here, and favoring slightly the estimates of potentialI. The GNP gap for 1976 can be decomposedinto severalparts. Closing the unemploymentrate gap of 2.8 pointswould increasetotal employment by 6.0 percentwhile expandingthe labor force by 2.9 percent.Employ- mentin the businesssector would rise 8.6 percentand averagehours would increase1.2 percent,for a total rise of 9.9 percentin (unweighted)hours. Togetherwith a 2.6 percentimprovement in productivityin the business sectorconventionally measured, these increaseswould close the 12.3 percent gap in 1976 businessoutput estimated using potential I. Withpoten- tial II, productivityin the businesssector would be 3.5 percenthigher than it was in fact, closingthe 13.3 percentbusiness output gap in that estimate.

POTENTIAL ESTIMATES COMPARED The new CEA estimatesof the 1976 level of potentialappear somewhat pessimistic,although they representa clear improvementover the former officialprojections, which were really made obsolete by last year'srevision of benchmarksin the GNP accounts.In shiftingto a 1972 base for the pricedeflator, the annualgrowth rate of real GNP was reducedby about 0.2 percentagepoint. This can accountfor a large part of the apparently excessivegrowth of the old series.Another important part of the difference must arise from the redefinitionof the unemploymentrate at potential fromthe old 4 percentrate to the 4.9 percentfor 1976 in the new estimates of the Councilof EconomicAdvisers. About half of the differencebetween the 1976 estimateof potentialI andCEA's revised one reflectsthe differencein the benchmarkyear, 1955. George L. Perry 43 If the CEA seriesis adjustedto the 1955 level in mine,their 1976 potential becomes $1,370.7 billion, or only 0.6 percentbelow potentialI and 1.2 percentbelow potential II. My estimateslend no supportto the pessimistic assessmentoffered by CEA that their own estimatesof potentialmay be $30 billiontoo high basedon an alternativereading of recentproductivity developments.21While their estimateof the 1976 gap is $99 billion and mineis $115 billionto $124 billion,their pessimistic assessment implies a gap of only $69 billion. FuturePotential. An even moreimportant difference with CEA comes in projectinggrowth rates of potential.From 1976 to 1980, CEA projects a growthrate of 31/2 percent a year, while the annual growth rate of potentialI is 3.88 percentand that of potentialII, 4.05 percent.Even the more conservativeprojection of potentialI for 1980 is some $40 billion ( 1972 prices) aboveCEA's. The slowdownin the growth of the capital stock and the impact of higherenergy prices are two reasonsgiven by the CEA for its pessimistic estimatesof recentand prospectivepotential GNP. Since the effect of the capitalstock on potentialproductivity is so hard to identify,it is impos- sibleto evaluatethis concern.Modest deviations from past growthrates in capital-outputratios may well have only negligible effects on potential productivity.The capital stock can indeed pose a barrierto reaching potentialif it turns out that industrialcapacity is inadequateto meet futuredemands and thus inflationaryprice pressuresdevelop while labor marketsare still slack. But that is a differentissue and one that lies well beyondthe scope of thispaper.22 I findit hardto see how high energyprices can affectproductivity and potentialoutput much. Energy is one input into productionprocesses. Eventually some adjustmentof production processes to conserve on energyis likely.But the effecton laborproductivity can hardlybe measur- able. As Okunhas calculated,on generousassumptions about how much laborcould be substitutedfor energyin responseto the pricerise of 1974, the productivitydent could not exceed$3 billion,or 0.2 percent.23 My own projectionsof a nearly 4 percentrate of potentialgrowth in

21. Economic Report of the President,January 1977, p. 55. 22. A recent analysis of this problem is made by Barry Bosworth in "Capacity Creationin Basic-MaterialsIndustries," BPEA, 2:1976, pp. 297-341. 23. Arthur M. Okun, "Unemployment and Output in 1974," BPEA, 2:1974, p. 503. 44 Brookings Papers on Economic Activity, 1:1977 the comingyears rest on the demographicfact of a maturinglabor force and what appearto me reasonablyconservative projections of otherparts of the puzzle.Table 14 summarizesthe ingredientsunderlying the projections of the potential I path and compares them with past intervals. Growthof both the potentiallabor force and employmentslows to a 2.1 percent annualrate in the 1976-81 period, a little slower than in the 1965-70 intervaland substantiallyslower than over 1970-76. Growthin businessemployment is sharplylower than in 1970-76 as the reduction in the armedforces is not assumedto continue.Average hours decline at the samerate as during1970-76, an assumptionI find hardto justifybut also hardto fault. The one thing we know about-demographic shifts- could bringwith it a slowerdecline. The favorabledevelopment is in output per (unweighted)hour. The changingemployment mix that was pro- jectedin table 8 resultsin a much smallerdifference between convention- ally measuredand weightedproductivity growth than has been true over the pastdecade. Giventhe estimatesbased on potentialI, reducingunemployment to 5 percentby reachingpotential output for the year 1981 would requirea sustained, strong expansion. Starting from 1976, actual employment wouldhave to rise by an averageof 3.2 percenta yearand real GNP would have to grow at an averagerate of 5.7 percent a year. Allowing for a gradualdeceleration of the rate of economicexpansion as potentialis approached,actual growthin GNP somewhatabove 6.0 percentwould be neededover most of the period.Should the path of potentialII turnout to be the rightdescription of the future,these same GNP growthrates would leaveunemployment at about5 1/2percent in 1981. There is no precedentin postwarU.S. economic performancefor the sustainedhigh rates of expansionprojected here. The expansionof 1961- 66 comesclosest. However, it startedwith a smallerGNP gap, and potentialwas growingless rapidlythan now projected.Furthermore, that expansion began in a period without inflation. It remains an open question whether,in comingyears, inflationwill deterpolicymakers from pursuing the rapid increasesin GNP that would make 1981 a year of 5 percent unemployment.The answerto that questionwill dependon many things: how inflationresponds to the expansion;how successfulthe government is with specific anti-inflationarypolicies; how much the price level is affected by an energy program and other economic developmentsor policies that are largely unrelatedto the level of unemploymentor the George L. Perry 45 Table 14. Profile of Changes in the Economy at Potential, Selected Intervals, 1955-81 Annual rate of growthin percent Projected Sector and economicmeasure 1955-60 1960-65 1965-70 1970-76 1976-81 Total economy Labor force 1.01 1.29 2.17 2.39 2.08 Employment 0.97 1.23 2.14 2.31 2.07 Real GNP 3.49 3.49 3.53 3.91 3.88 Businesssector Employment 1.21 1.11 2.30 2.87 2.07 Total hours 0.95 0.84 1.41 2.41 1.53 Output 3.62 3.42 3.64 4.34 3.92 Outputper hour 2.65 2.56 2.19 1.96 2.35 Outputper weightedhour 2.79 2.79 2.67 2.48 2.49 Source: Author's estimates based on potential I path. rate of expansion; and how policymakers balance unemployment targets against price stability or other economic goals. Also, if policies to provide public jobs directly are pursued vigorously even after unemployment is substantially reduced from present levels, then the historical link between unemployment and GNP may be stretched, producing a somewhat greater decline in unemployment per dollar of GNP than has been estimated here. A path for GNP that even approached the one mapped here for achieving 5 percent unemployment by 1981 would require a large expansion of the capital stock in coming years. And achieving this expansion would mean a rapid rise in business investment. That, in turn, would help spur the increase in aggregate demand required and would permit, and eventually require, a reduction in the present full-employment deficit of the federal budget.

APPENDIX

AlternativeParticipation Equations

TABLE A-1 presents the estimates of cyclical and trend effects on participation rates from several equations. Cyclical effects are expressed by the 46 Brookings Papers on Economic Activity, 1:1977 sum of the estimatedcoefficients on the weighted unemploymentrate. Trend effectsare expressedby the sum of the time-trendcoefficients and thusgive the estimatedtrend growth in participationrates for recentyears. The equationsare all of the form of equation1 in the text. The estimates differaccording to the periodover which the equationswere estimatedand the yearin whichthe timetrends start.

Table A-1. Effects on Labor Force Participation Rates of Equations with AlternativeEstimation Periods and Time Trends, by Age-Sex Groups

Effect, Age group period of estimation, and startinigyear of 65 and time trend 16-19 20-24 25-34 35-44 45-54 55-64 over Males Cyclicaleffect (sum of unemploymentcoefficients) 1949-76 1967 -2.99 -1.79a -0.12 ...... -2.06k 1970 -3.78 -1.79a -0.11 ...... -2.06a 1954-76 1967 -2.95 -1.28 -0.07 ...... -2.89 1970 -4.31 -1.59 ...... -2.20 1949-73 1967 -3.39 -2.00a -0.12 ...... -2.24 1970 - 3.95 -2.OOa -0.11 ...... -1.24 Trendeffect (sum of time-trendcoefficients)b 1949-76 1967 1.18 ... -0.19 -0.17 -0.40 -1.37 ... 1970 2.35 ... -0.21 -0.23 -0.51 -1.78 ... 1954-76 1967 1.30 0.10 -0.18 -0.16 -0.38 -1.30 -2.04 1970 2.65 0.46 -0.25 -0.17 -0.50 -1.69 -3.11 1949-73 1967 0.49 ... -0.23 -0.12 -0.35 -1.20 -2.13 1970 1.78 ... -0.30 -0.21 -0.47 -1.77 -3.87 Standard errorsb 1949-76 1967 2.8 1.4 0.3 0.2 0.2 0.7 2.5 1970 2.6 1.4 0.3 0.1 0.2 0.7 2.5 1954-76 1967 2.9 0.9 0.2 0.2 0.2 0.7 2.5 1970 2,6 0.9 0.2 0.1 0.2 0.5 2.6 1949-73 1967 2.7 1.4 0.3 0.2 0.2 0.7 2.5 1970 2.7 1.4 0.3 0.1 0.2 0.7 2.6 George L. Perry 47 Table A-1 (Continued)

Effect, ge group periodof estimation, and startingyear of 65 and time trend 16-19 20-24 25-34 35-44 45-54 55-64 over Females Cyclicaleffect (sumof unemploymenitcoefficients) 1949-76 1967 -3.34 -2.42 -2.20 -1.81 -0.99 -0.94 1970 -4.46 -2.05 -2.72 -1.92 -0.95 -0.83 ... 1954-76 1967 -3.32 -2.65 -1.97 -1.67 -0.51 -0.89 ... 1970 -4.63 -3.32 -2.58 -1.93 -0.43 -0.56 ... 1949-73 1967 -3.77 -2.88 -3.28 -2.22 -1.13 -1.18 ... 1970 -4.69 -2.20 -3.20 -2.24 -1.10 -1.13 ... Trendeffect (sumof time-trenidcoefficients)b 1949-76 1967 3.09 2.89 4.12 2.33 0.49 -0.44 -2.36 1970 4.57 3.26 5.25 2.63 0.16 -0.94 -3.49 1954-76 1967 3.13 2.87 4.13 2.43 0.45 -0.27 -2.27 1970 4.55 3.69 5.10 2.88 0.05 -1.14 -2.97 1949-73 1967 2.76 2.70 3.61 1.83 0.30 -0.60 -2.26 1970 4.61 3.10 4.75 1.89 0.38 -1.55 -3.53 Stanzdarderrorsb 1949-76 1967 2.6 1.7 2.0 1.4 1.8 1.9 4.3 1970 2.7 2.0 2.1 1.4 1.8 2.0 4.4 1954-76 1967 2.8 1.5 1.7 1.2 1.4 1.6 3.3 1970 2.9 1.6 1.8 1.3 1.3 1.5 3.0 1949-73 1967 2.7 1.7 1.7 1.2 1.9 2.0 4.6 1970 2.9 2.1 2.1 1.3 1.9 2.0 4.6 Source: Estimated by author. a. Neither time trend appeared in this equation. b. The results were multiplied by 100. Commentsand Discussion

MichaelL. Wachter:George Perry has updatedhis 1971 paper on potential output, concluding that the traditionalmacroeconomic supply model is still reliable and that the economy has the potential for full- speed-aheadeconomic growth between 1977 and 1981. Methodologi- cally, Perrypresents a viable, coherent,and consistentmodel of what I view to be the "optimistic"-perhapshighly optimistic-story of current andnear-term supply developments. My approachto the supplyside inter- prets the data in a differentframework and arrivesat significantlylower estimatesof potentialoutput. In this commentI utilizemy frameworkfor a critiqueof Perry'smethodology and to indicate the various areas in whichhe maybe overstatingpotential output. Most of my commentswill be directedat the labor supply equations, which are, as Perry notes, a pivotal componentof his potential output series. I have long been surprisedthat Perry, having devised the initial adjustmentof the unemploymentrate for changesin the labor force mix, would remain a proponentof the large GNP gap. The old measure of the gap was built on the foundationof a 4 percent equilibriumunem- ploymentrate, and the Perryindex clearlyexhibits a nonacceleratingin- flationrate of unemploymentconsiderably higher than 4 percent.If the benchmark,noninflationary rate is high,where are the extraunutilized re- sourcesnecessary to generatea large GNP gap?Part of the answerto this problemis that Perry has hidden away a large group of "discouraged workers."These workerswill flood into the labor market,at unchanged real and nominalwage rates, to providethe upwardratchet for potential output.Indeed, for some age-sex groups,the coefficientsof table 1 pre- dicting new additionsto the labor force per net job created suggest a delugeof new entrantsor reentrants. 48 George L. Perry 49 My own resultssuggest a differentinterpretation of the participation rates; and, Perry'snew calculationsnotwithstanding, I believe that the evidencefor a sizable discouraged-workercomponent is shaky. Perry's significantand quantitatively important unemployment coefficient in table 1 is highlydependent on his specificationof the participationmodel. In particular,Perry's ability to discountthe inflationvariable as explaining participationrather than unemploymentis affectedby his use of a split- trendvariable broken in 1967 or 1970. My earlierresults suggested that the discouraged-workermodel began to generateperverse predictions in 1969-70, a period in which unemploymentincreased and participation rates rose significantly.The split trend in 1967 or 1970 allows Perry to relabelthis seeminglyperverse rise in participationrates as partof the new sociologicalattitudes of youngerwomen towardmarket work. Since inflation rates also acceleratedafter 1970, these two variablesare highly collinear.In straighthorse races between unemploymentand inflation rates,without the broken trend, the latter does considerablybetter than it does in the resultsreported in table4. Using the inflationvariable in place of the discouraged-workereffect suggestsseveral changes in the interpretationof the potential-outputmea- sure. First, unless the inflationrate continuesto trend upward,most of the cyclicallysensitive or "inflation-sensitive"workers are alreadyin the laborforce. Althoughthe labor force will continueto grow rapidlyover the next few years because of underlyingpopulation growth in the rele- vant age groups,there should not be a new "bonus"of newly encouraged workersas unemploymentfalls to its noninflationarylevel. Second,the implicationsof the presenceof inflation-sensitiveworkers are differentfrom those of discouragedworkers. The latter are a pure bonusin thatthey enterthe laborforce regardless of wage andprice move- ments.The formerare a part of the supplypicture only at a cost in inflation. An exogenouspotential output, measured at zero inflation-in the Perrytradition-would excludethese workers. Third, althoughthe empiricalresults are ambiguous,many cyclically sensitiveor inflation-sensitiveworkers are likely to be only temporary additionsto the labor force at full employmentwith 6 percentinflation. As Mincer originallywarned us, the discouraged-workereffect may be measuringthe "timing"of part-timeworkers. If some workerswant to workpart of the time, they have an obviousincentive to coordinatetheir entry into the labor market and their employmentwith periods of low 50 Brookings Papers on Economic Activity, 1:1977 unemploymentrates (and high discountedreal wages). Empirically,there is considerableevidence of a large contingentof workerswho work part of the timeand frequently move into andout of the laborforce. The better or more rationalthe timingof these workers,the greaterthe appearance of a large "discouragedworker" contingent. If the economy managesto stay at full employment,these workerswill eventuallydrift out of the labor force, reducingpotential output. This timing effect may also be relevantto inflation-sensitiveworkers to the extentthat they are respond- ing to short-termmoney illusion. I do not believe that these workersare all temporaryadditions, but it is highly optimisticto believe, at the other extreme,that they are all permanentadditions. In short, I would favor excludingthe discouraged-workercontingent from an exogenousmeasure of potentialoutput. It is unclearthat these workersare cyclicallyrather than inflationsensitive and that they will remain in the labor force if the economyis at full employmentfor any extended period of time. Removingthis group would considerablyreduce Perry'spotential output measure for 1976. An additionalissue in the equationsfor labor force participationcon- cernsthe role of relativewages. Here a data problemis the main culprit in Perry'sinsignificant and incorrectcoefficients. Although Perry uses the aggregatewage as the basis for the w/w* construct,the desiredvariable is an unavailableage-specific wage rate. Apparently,it is largely after 1970 that age-specificand aggregatewages diverge in any meaningful way. In particular,aggregate real wages increase dramaticallyin 1971 and 1972, but not those for young workers.The sharp increase in the relative percentageof young workers reduced their relative wage as it increasedtheir relativeunemployment rates. My argumenthas been that the increasein participationrates for young females and the concomitant decreasein fertilityrates is relatedto their (relatively) adverselabor market experiences.This has been confirmedby the dataof the 1970s although it doesnot appearin the aggregatedata on wagerates. Fluctuationsin the laborforce due to changesin relativewages among cohortsare intermediate-run swings that are similar in some respectsto the Kuznetscycle. Hence they are not quantitativelyimportant for Perry's short-runprognosis. They do have significantimplications, however, for extrapolatingthe growthof the labor force into the 1980s. In particular, these models suggestthat labor force growthwould be considerablyless robustthan is suggestedby using the Perrymodel to forecastthe 1980s. George L. Perry 51 Perry'strend variable mechanically extrapolates into the futuresome- thing like the very rapid recent increasesin participationrates among young females.In the contextof an overall demographicmodel, this implies a furthersignificant drop in fertilityand school-enrollmentrates for youngpeople. I believethat this outcomeis unlikely.As the currentbaby- boom cohortages, the high level of participationrates for young women todayis likelyto be reflectedin higherparticipation rates for olderwomen in the future.It is much more problematical,however, to predictthat the next cohortof young women will have an even higherparticipation rate than the currentcohort. Again, these considerationssuggest that Perry'slabor force projections are too high. In addition,they suggestthat potentialoutput should not be treatedas an exogenousvariable. In the short and long runs, the supply constraintof society is responsiveto fluctuationsin economicvariables. Besidesoverstating labor force growth,Perry may, I believe, overstate the noninflationaryrates of employmentfor the differentage-sex groups. My commentshere have been based on the model I outlinedin a previous paper (BPEA, 1:1976), and they suggestan increasein the steady-state unemploymentrate of approximatelyone-half percent (on aggregateunemployment)beyond that suggestedby Perry.In addition,if that model is correct,Perry's wage-weighting scheme is also too optimistic.The larg- est cohortswill sufferan adverserelative-wage adjustment. Hence, Perry is alwaysgiving the largestgroups too high a relativewage rate. This is especiallya problemfor near-termprojections as the boom-babiesage and enterthe prime-age categories. In his discussionof productivityPerry seems appropriately cautious on the questionof the long-termproductivity trend. This remainsan area of substantialuncertainty. His productivityestimates, however, specifically ignorethe potentialproblems that he identifies.Of particularconcern to me is the continuingshift betweenunemployment rates and capacityuti- lizationrates as measuredby eitherthe Whartonor the FederalReserve Board index. In all cases, the shift indicatesthat the noninflationaryun- employmentrate may be above 5.5 percent (at least in the short run) or thatthe capital-stocknumbers are overstated,or both. The qualityof the data on capacityand capital stock is open to some question;however, it seemsunlikely that errorsin thesevariables would makethis problemdis- appear entirely. In other words, Perry's estimates representthe upper boundof the supplyconstraints on the capitalstock. 52 Brookings Papers on Economic Activity, 1:1977 In general,Perry is consistentin developingan upper bound for potential output. His calculationsmay well turn out to be correct, but a growthpolicy basedon his supplyseries has majorinflationary risks.

Otto Eckstein:George Perry's paper carefully updates his earlierstudy of the full-employmentlabor force and potentialGNP, and exploresecono- metricallythe changingparticipation rates for the major groups in the population.He confirmsthat the participationrate of women has accel- eratedand is rooted in sociologicalfundamentals. He also finds a break in the time trendsfor adultmen, whose participationrate is falling.While he offerslittle explanationof this phenomenon,one can easilyturn to such factors as changes in the social security,unemployment insurance, and welfaresystems, early retirement provisions in collectivebargaining agree- ments,perhaps the regionaldistribution of older workersas comparedto changing job opportunities,perhaps the increasing competition from womenfor certaincategories of jobs, perhapseven a decline in the work ethic. As medicalknowledge of heart disease and other degenerativedis- eases improves,we may live to regretthe policy decisionsthat have lowered the retirementage and that have hurtthe participationrates of male workersincreasingly from age 25 on. But there are two other importantquestions raised more directlyby the Perrypaper: First, shouldwe acceptthe Perryestimate of a potential growthrate near 4 percentfor the next five years,contrary to the conclusions of variousother studiesincluding those by Roger Brinnerfor DRI and by the Council of Economic Advisers in their last annual report? Second,for the kind of economyin which we now live, what really is the pertinenceof the concept of potential output? I shall deal with these brieflyin turn. Perry concludes that the probable annual growth rate for potential GNP for the period 1976-81 is 3.9 percent.This is roughlyhalf a point higherthan the CEA or DRI conclusions.When Brookings disagrees with the CEA andDRI, we hadbetter take a look at the reasons. The table below juxtaposesthe componentsof the growthof potential GNP for the Perry (Brookings) and Brinner(DRI) studies.It shows the sourceof differencebetween the two projections.There is no substantive disagreementin the analysisof the growthin the full-employmentlabor force or in potentialoutput for the historicalperiod, althoughthere are differencesin the underlyingmethods. Brinner uses somewhatdifferent George L. Perry 53 breaksin the age compositionand uses unweightedunemployment rates. His formulasfor the cyclical element in participationrates are of the standardtype, as are Perry's.Differences in definitionand slightlydiffer- ent periodslead to differencesin the growthrates of 0.1 point. For the periodsof the individualstudies shown in the table, the potentialgrowth rates for the labor force are very close, althoughover the specific 1976- 81 interval that Perry reports on, the Brinner projectionwould show slightlylower growth.However, the forecast horizonreveals the impact of the differentmethodologies plus alternativeassessments of the outlook for productivitygrowth. Perry study Brinnerstudy Componentof growth 1960-76 1976-81 1963-73 1973-80 Compoutndannual rates of growth Laborforce 2.0 2.1 2.1 2.1 RealGNP 3.7 3.9 3.8 3.4 ImplicitGNP per person 1 .6 1 .8 Total inputs ...... 2.8 2.3 Total factor productivity ...... 1 .0 1 .0 Perrybases his analysison the labor-productivitytrend, while Brinner uses the aggregate-production-functionapproach. There may have been a time when the two approachescame to about the same conclusion,and both are of long standing.The projectionsin a 1960 study for the Joint Economic Committee,by James W. Knowles, The Potential Economic Growth in the United States, used a Cobb-Douglas production function. This was one of the first analyses of the problem, and its projections provedto be correctfor the 1960s. While the two approachesmay have given similaranswers in the past, they cannotbe expectedto do so now. Growthof the capital stock has slowed severely-from 3.9 to 2.9 percent-because of the liquiditycrisis in 1973-74 and the subsequentdeep recession.With the Cobb-Douglasfactor-share weights, this decline cuts the growthof potentialoutput by a thirdof a point. Could such a dramaticreduction in the growthof capitalhave no impact on the growthof potential?Only a strangeeconomic theory would reach that conclusion.Perhaps technology is really so disembodiedthat the Cobb-Douglasfunction inadequately represents the role of capital.But thatthere be no relationbetween capital and outputis impossible!And yet that is what an analysis of potential output based on nothing but labor productivityand labor supply would require. 54 Brookings Papers on Economic Activity, 1:1977 Furthermore,the labor-productivityapproach breaks down statistically for recentyears, as Perryshows. He finds a breakin the time trendsand employs a lower time trend beginningwith 1969. That trend creates a slowdownof 0.3 to 0.4 percentfrom that date onward.Further, it is nec- essaryto dummyout 1974, which admittedlywas a very strangeyear for productivitybecause business was operatingon a false set of expectations. But with 1974 dropped,it is difficultto use 1975, the year afterthe liquid- ity crisisforced business to cut employmentby 2.4 millionpeople between September1974 andApril 1975. The aggregate-production-functionapproach runs into some of the sameproblems since laboris still the principalfactor of production.But it is possible to fit equationsthat treat 1974 as just anotheryear with an error,and that identifya trendin total-factorproductivity that is reason- ably constant.To the extent that there is any doubt about it, the factor- productivitytrend gives some statisticalhints of a small slowdownin the early 1970s. The Brinner formulationassumes the factor productivity trendto be constant.If that trendwere assumedto have slowed down by 0.1 or 0.2 percent (or more), which some equationforms and intervals yield, the projectedgrowth of potentialwould be even lowerthan the estimatesof 3.4 percentthat he advances. The concept of potential GNP is the purest Keynesian macroeco- nomics that can still be found. Originally,potential GNP was seen as the gauge for identifyingthe gap of underutilizedresources, and was prin- cipally the baseline for calculatingthe "inflationarygap" advancedby Keynesin How to Pay for the War and still shownin basic textbooksas a methodof inflationanalysis. For thatpurpose, potential GNP mustnowa- days be consideredthoroughly inadequate because it has nothingto say about the cost side or the dynamicsof inflation.Even as a measureof demand,it focusessolely on the aggregateunemployment rate, completely leavingaside the morepressing problems of capacitybottlenecks, resource shortages,or even skill shortageswithin the labor market.To arguefor any particulareconomic policy because there is a full-employmentgap, as indicatedby potentialGNP, nowadaysreally is an error.To assessthe inflation implicationsof any particularpolicy, one must use far more sophisticatedtechniques-in my view, a complete,detailed model in which the gap is just anothervariable. While too crude a measurefor inflationanalysis, potential GNP remains an essentialingredient in the most precise method for projecting aggregateunemployment. It remainsas true today as it was a decadeago George L. Perry 55 that Okun's law is the best predictorof aggregateunemployment, and that aggregateunemployment is the dominantexplanatory variable for most sectoralunemployment rates. Thus potentialGNP, along with lots of other information,is needed to formulatemacro policies designedto reach specificunemployment goals. Further,the growthof potential GNP is a determinantof the overall scale of the economy.In long-termanalysis of such demandsas those for housing or automobiles,the aggregategrowth of real purchasingpower, along with demographicand other variables,determines the long-term trend values. Dependingupon income elasticities,these trendvalues will varysystematically with the trendof potentialGNP. For these reasons, a carefulformulation of potentialGNP remainsof importanceeven after the numerouslimitations of this still very aggrega- tive concept are recognized.It is my belief that we will obtainbetter economic performanceif we resolve our doubts aboutpotential GNP on the side of caution.

Peter K. Clark:There are some similaritiesand manydifferences between Perry'smethodology for obtainingan estimateof potentialGNP and the techniqueused in obtainingthe revisedseries on potentialGNP published in the 1977 Economic Report of the President. The most striking similar- ity is the estimateof 4.9 percentas the overallunemployment rate in 1976 equivalentto 4.0 percentin 1955. This similarityis less surprisingwhen one realizesthat neither Perry's technique nor the techniqueimplicit in the council estimateuses a structuralmodel of unemployment;instead, both estimatesare an empiricaldescription of the rising proportionof young personsin the laborforce, and risingunemployment rates for theseyoung people relativeto the unemploymentrate for adults. One of the differencesin Perry'sresults is the strongcyclical sensitivity he finds in the labor force. For example,in 1976, Perry estimates,if the unemploymentrate had been 2.8 points lower, the labor force would have been 2.8 million larger.In general,if employmentincreases by ten persons,between four and five additionalworkers enter the labor force. The reason for this large increase is the strong cyclical dependenceof laborforce participationrates for adultwomen (aged 25-64). The council's revisedestimates of potentialGNP assume an increasein the labor force of about 0.4 percent for every percent decreasein the unemploymentrate. That implies an increaseof only 1.1 million in the labor force for a dropof 2.8 pointsin the unemploymentrate. 56 Brookings Papers on Economic Activity, 1:1977 Other differencesbetween Perry'smethodology and the council'sin- clude the latter'suse of quarterlydata, which shouldbe more accuratein capturingthe effects of cyclical variabilitythan Perry'sannual data, and its use of the capitalstock and total-factorproductivity rather than labor productivity.Instead of belaboringthese differences,however, it may be more instructiveto look at some alternativemethodologies that were used neitherby Perrynor by the councilfor a resolutionof the conflictbetween the two potentialGNP series. As a roughcheck on the level of potentialGNP in 1976, one may calculate the percentagechange in real GNP (Q) associatedwith a given percentage-pointchange in the unemploymentrate by estimatingan equation of the form

(1) Ut -Ut- a + b (Q) + et, where u = the overallunemployment rate in percent AQ/Q = percentchange in gross nationalproduct in 1972 dollars. This is, of course, the firstof the three equationsused by Okun in his originalestimate of potentialGNP. Using annualdata for 1953-76, the followingregression results were obtained:

Aut= 1.45 - 0.40 Q (0.18) (0.04) Q/t -R = 0.78; Durbin-Watson== 2.37. The figuresin parenthesesare standarderrors. This equationsays that a 1 percent increase in real GNP reduces unemployment0.4 percentage point below what it would have been otherwise.The constantterm 1.45 is an estimateof how much the unemploymentrate will rise duringa year in which no real growthoccurs. The implied "Okun'slaw" multiplieris 2.5, much less than the 3.1 reportedby Perry.Regressions excluding the years 1974-76 produceOkun's law multipliersthat are even smaller. If a lagged value of percentagechange in real GNP is enteredin the equationin orderto obtain a "longer-run"reaction of unemploymentto output,the resultsare a lowermultiplier:

Aut = 1.73 - 0.40 (Q) - 0.08 ( Q (0.04) Q t (0.04) Q t-l R2 = 0.80; Durbin-Watson 2.40; annual data, 1953-1976. The impliedOkun's law multiplieris 2.1. George L. Perry 57 The same equationusing quarterlydata providesa similarpicture:

ALtt= 0.38 - 0.24 ,_Q_ - 0.18 (Q (0.04) (0.02) Q t (0.02) Q t-/ PI = 0.77; Durbin-Watson=1.89; p = 0.23; quarterlydata, 1953:2-1976:4.

Aut = 0.45 - 0.25 0.10 ( Q) - 0.08 ( Q) (0.04) (0.02) Qt (0.01) \Q t (0.0l) Q t-2

- 0.05 Q - 0.03 tQ (0.005)\ Q/ t- (0.003) Q t-4 R1 0.75; Durbin-Watson= 1.89; p-0.27; quarterlydata, 1953:2-1976:4. The last four regressioncoefficients are constrainedto lie on a straight line. The impliedOkun's law multipliersare 2.4 and 2.0, respectively.Of course, a considerablyhigher multipliercould have been obtained by using quarterlydata and no lagged values for output changes, but this procedurehas serious specificationbias due to the lagged response of laborinput to output. From these regressions,which are not influencedby a prior estimate of potentialGNP, I concludethat a decreaseof one percentagepoint in the unemploymentrate would increasereal GNP 2.0 to 2.5 percent.Ap- plied to an "unemploymentgap" of 2.8 percentagepoints in 1976, this multiplieryields a potential GNP of $1,335 billion to $1,353 billion (1972 prices), somewhatless than the councilestimate of $1,364 billion and muchless than the lowerof the two Perryestimates of $1,379 billion. If the Okun'slaw multiplieris 2.5 or less, how did Perry get 3.1? The answeris clear: Perry'sestimate of the reactionof unemploymentto output is biaseddownward by his high estimateof potentialoutput. The equationin Perry'stable 13 is:

u-u = a,(Q Q) + a2D74.

The higherthe level of potential (Q), the lower the regressioncoefficient a,, and the higher1 /a,, the Okun'slaw multiplier.Although estimation of the multiplierusing this form of Okun'slaw usually results in a slightly largernumber, the increaseis not large enough to change the 2.5 result whenlagged values of outputare included in the equation. Aside fromthe roughOkun's law estimatesof 1976 potential($1,335- $1,353 billion in 1972 prices) just discussed,there is other evidencethat the level of productivitymay have shifteddownward in 1974. The follow- 58 Brookings Papers on Economic Activity, 1:1977 ing equationestimates the trendand cycle in privatenonfarm productivity from 1953 to 1976:

LP = -0.55 - 0.0146URMA + 0.OO8OURMA(-1) + 0.00684TIME (0.01) (0.002) (0.002) (0.0002) - 0.00213T1 - 0.028D1 - 0.024D2 - 0.019D3, (0.0003) (0.007) (0.008) (0.010) TV = 0.998; Durbin-Watson= 1.53; p = 0.59; quarterlydata, 1953:2 to 1976:4. where LP = logarithmicindex of laborproductivity in theprivate nonfarm sector,'all persons URMA = unemploymentrate of men 25-54, adjustedfor the sampling changein 1967 TIME -time trend, 1, 2, 3, 4, ... Tl = 0 through 1966:4, then 1, 2, 3, 4 ... DI = 1974 dummy,equals 1.0 in 1974,zero otherwise D2 = 1975dummy D3 1976dummy. The magnitudeof the D2 and D3 coefficientssupports the possibilitythat a 2 percentdrop in the level of productivityhas been maintainedthrough 1976. Alternativeestimates using capitalinputs and total factor productivity strengthenthis conclusion,since capital utilizationhas rebounded faster in 1975 and 1976 than has labor utilization.Although an extraor- dinary cyclical fall in productivityin 1974 is easy to understand,the persistenceof this gap throughoutthe recoveryyears of 1975 and 1976 casts considerabledoubt on the hypothesisthat the post-1973 productivity shortfallis only temporary. These two empiricalchecks tend to confirmthe council'srevised estimate for potential GNP for 1976 and to reject the figure proposed by Perry.As for the future,since both Perry'sprojections and ones based on the council'srevised figures are trend extrapolations,neither provides a very reliable base for a forecast.However, the bad productivityperfor- mance since 1966 and the steep productivitydecline in 1973-74 caution one about future productivitygains. The council's current (unofficial) projectionof outputat 4.8 percentunemployment in 1981 of $1,623 billion (1972 prices), which assumesthat the post-1973 productivityshort- fall will be eliminated,represents a modestlyoptimistic view of growthin the next five years. George L. Perry 59 George L. Perry:I disagreefundamentally with Wachteron the role of wage variablesas opposedto unemploymentin the estimationof the behavior of participationrates. One can think of unemploymentequa- tions as reducedforms of relationsthat includethe real wage that matters to workers:cyclically, this relevantreal wage changeswith explicitwage differenceslike those discussedby Okun and Vromanin connectionwith cyclical upgrading,and with indirectdifferences that show up in conve- nience and other nonwage aspects of the job. None of this gets measuredby availablewage series, but it is relatedto unemployment.I also disagree basically with Wachter'spreference for inflation variables in place of trends.I can imagineroom for both and I used both. The trend variablewas used to capturesomething quite differentfrom inflation,a complexsociological phenomenon that economistsought to take as given and not pretendto explainin termsof a few simplemacroeconomic variables. The women'smovement is not simplya responseto unanticipated inflation. Nor do I believe models that assert that the participationinduced by low unemploymentis temporary.The biggestcyclical response is found for teenagers.Are they just reallocatingtoward the presenttheir lifetime quota of work, planningto reduceit correspondinglyat some later date? This view is not persuasivefor any labor force group. Indeed, quite the opposite seems to be the case. Work is habit forming and being drawn into the work force in the present apparentlyraises the probabilityof being in it in the future.Thus I see no reasonnot to treat estimatedpar- ticipationeffects as permanentand as a componentof potential output measures.

GeneralDiscussion

Some participantsdiscussed the possible inflationaryconsequences of operatingat the unemploymentrate underlyingPerry's potential output. RobertJ. Gordonreasoned that the 1955 benchmarkperiod was one of acceleratinginflation; and he noted that for more recentyears, both he and MichaelWachter had obtainedhigher estimates of the unemployment rate associatedwith nonacceleratinginflation. Charles Holt questioned the use of a constantweighted unemployment rate as a standardfor de- finingpotential output. He consideredboth the choice of efficiencyunits 60 Brookin2sPapers on Economic Activity, 1:1977 as a weightingmeasure for differentgroups of unemployedand the use of 3.32 percent as a referencelevel for weighted unemploymentas quite arbitrary.He felt that, ideally, an explicit theory of wage and price dy- namicsshould be incorporatedin derivingthe weightsappropriate to each group;and this theory,in turn,would providea measureof the noninfla- tionaryunemployment level to link with the potential-outputanalysis. Lawrence Klein felt that the omission of an explicit treatmentof energy and resource limitationscast doubt upon the estimates of the long-runtrend. Simulationswith large-scalemacroeconomic and input- outputmodels indicate that high prices for oil and otherraw materials and high energy requirementslead to large trade deficits and high inflation ratesalong GNP pathscomparable to Perry'spotential. Although cyclical recoverymight allow temporarygrowth at rates greaterthan potential, the long-runpath would have to be adjusteddownward unless specific policieswere adopted to removebottlenecks. Franco Modigliani disagreed that tradedeficits would be a hindrancein attainingpotential. Either the OPECcountries would buy goods from the United Stateswith the money they earnedfrom oil or they would sell us the oil on credit. Modiglianicriticized Otto Eckstein's argumentthat the productivity trendmust have declinedbecause of a lower rate of capitalformation as simply a blank assertion.Not only was the relationbetween capital and labor productivityuncertain, but the future growth of capital was not yet known.Robert Hall conjecturedthat capitalmight not be closely re- lated to movementsin laborproductivity, supporting Perry's decision not to incorporatecapital as an explicit determinantof potential output. Production-functionestimates always make use of additionalevidence on capitalincome, and althoughit is commonpractice to attributeall profits to physicalcapital, intangible capital might actually be a substantialpart of the story. However, Eckstein cautioned that a bias understatingthe importanceof capitalstemmed from considering all technologicalprogress as disembodied.