The Phillips curve. Back o he 60s? Olivier Blanchard The behavior of inflaion since he crisis appears puzzling o many. To cie Paul Krugman (2015): The acceleraionis docrine ha has dominaed economic discussion of inflaion and unemploymen for 40 years has fallen fla. If inflaion had responded o he Grea Recession and afermah he way i did in previous slumps, we would be deep in deflaion by now: we aren. Wih his in mind, his paper reexamines he behavior of inflaion and unemploymen. I reaches four conclusions: The U.S. Phillips curve is alive and well (or a leas as well as i has been in he pas). Inflaion expecaions however have become seadily more anchored, leading a relaion beween he unemploymen rae and he level of inflaion raher han he change in inflaion. In his sense, he relaion resembles more he Phillips curve of he 1960s han he acceleraionis Phillips curve of he laer period. The slope of he Phillips curve, i.e. he effec of he unemploymen rae on inflaion given expeced inflaion, has subsanially declined. Bu he decline daes back o he 1980s raher han o he crisis. There is no furher evidence of a decline during he crisis. The sandard error of he residual in he relaion is large, especially in comparison o he low level of inflaion. Each of he las hree conclusions presens challenges for he conduc of moneary policy. Wisdom learned from he experience of he 1960s and laer will be needed.
A brief lieraure review. This paper exends Blanchard, Cerui, and Summers (BCS in wha follows, 2015), where we examined he evoluion of he relaion beween inflaion and unemploymen for 20 advanced economies. Tha paper builds in urn on an IMF World Economic Chaper (2013) on he same opic. Our conclusions mosly coincide wih he empirical conclusions of he wo closes papers we know of on his opic, Ball and Mazumder (2011), and Kiley (2015) (which includes a review of oher papers). An economeric exercise. Since he esimaion of he U.S. Phillips curve by Samuelson and Solow, macroeconomiss have learned, ofen painfully, ha, while low unemploymen creaes inflaion pressure, he form of he relaion can change and has changed over ime. To examine is evoluion, we esimaed in BCS he following specificaion: * e ( u u ) (1 ) * 1 m * e 1 where is headline CPI inflaion (defined as quarerly inflaion, annualized), u is he unemploymen rae, * u is he naural rae, average of he las four quarerly inflaion raes, inflaion, and λ, θ, μ, β, α, and e is long-erm inflaion expecaions, 1 m * is he is impor price inflaion relaive o headline * u follow consrained random walks. The firs equaion specifies he Phillips curve. Inflaion depends boh on expeced long erm inflaion and on pas inflaion. The coefficien on pas inflaion reflecs he dependence of shor erm inflaion expecaions on pas inflaion as well as he direc effecs of pas inflaion on
curren inflaion. Inflaion also depends on he deviaion of he unemploymen rae from he naural rae, as well as on he relaive price of impors. The second equaion (which was no esimaed in he BCS paper) capures he dependence of long erm expeced inflaion on lagged inflaion. To capure he evoluion of he wo relaions over ime, he slope of he Phillips curve, he coefficiens on long erm inflaion expecaions in he Phillips curve, and on lagged inflaion in he expecaion equaion, he inercep of he expecaion equaion, and he naural rae of unemploymen, are allowed o follow random walks. (Esimaion is done using quarerly daa since 1960. Daa sources, and deails of esimaion for he firs equaion, are given in BCS.) The main resuls are presened in he hree figures below (For lack of space, resuls abou he evoluion of he naural rae are no presened here. The daa sugges a slow decline in he naural rae by abou 1 percenage poin since he early 1980s.) Figure 1 shows he evoluion of λ, he weigh of long erm expecaions in he Phillips curve. I shows how, afer going down in he 1970s, i has seadily gone up since he mid-1980s, and is now close o one. Equivalenly, he weigh of pas inflaion, 1-λ, has seadily decreased over ime. The anchoring of expecaions (λ) 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 61 65 69 73 77 81 85 89 93 97 01 05 09 13
Figure 2 shows he evoluion of β, he coefficien reflecing he effec of pas inflaion on long erm expeced inflaion. Afer increasing in he 1970s, i decreased in he 1980s, and has been close o zero since he lae 1980s. The anchoring of long erm expecaions (β) 0.3 0.25 0.2 0.15 0.1 0.05 0 61-0.05 65 69 73 77 81 85 89 93 97 01 05 09 13 Puing Figures 1 and 2 ogeher sugges ha inflaion now depends mosly on long erm expeced inflaion raher han pas inflaion, and ha long erm expeced inflaion in urn depends lile on pas inflaion. This implies ha he Phillips curve relaion is now very close o a level-level relaion, wih he level of he inflaion rae relaive o sable long-erm expeced inflaion depending on he level of he unemploymen rae. Figure 3 shows he evoluion of θ, he slope of he Phillips curve. The slope increased from he 1960s unil he lae 1970s, hen seadily decreased unil he lae 1980s and has remained roughly consan and low since hen. 1.2 The decrease in he slope of he Phillips curve (θ) 1 0.8 0.6 0.4 0.2 0 61 65 69 73 77 81 85 89 93 97 01 05 09 13
There is no evidence ha he slope has decreased furher in he crisis. Given expeced inflaion, a decrease in he unemploymen rae led o a decrease in inflaion of 0.7% in mid-1970s. The effec is now closer o 0.2%. Various explanaions have been offered for his evoluion. The mos convincing is ha, as he level of inflaion has decreased, wages and prices are changed less ofen, leading o a smaller response of inflaion o labor marke condiions. (In he Calvo formalizaion of price sickiness for example, he slope coefficien is roughly proporional o p 2, where p is he probabiliy ha a price will be changed in a given period.) The las relevan resul is ha he fi of he relaion remains fairly poor. The sandard deviaion of he residual is roughly equal o 1% (a an annual rae) oday, a large value relaive o an inflaion rae around 1-2%. This suggess ha he US economy is far from saisfying he ``divine coincidence, he condiion ha keeping inflaion consan delivers he bes unemploymen rae policy can deliver. Resuls vary slighly, depending on he exac choice of variables and he exac specificaion. Some specificaions, using differen measures of inflaion, give a slighly larger slope, and a slighly lower value for θ (see Ball, and Kiley). Bu he hree evoluions shown in he previous figures appear robus. They have imporan implicaions for he conduc of moneary policy. The end of he acceleraionis curse? One of he mos dramaic implicaions of he acceleraionis Phillips is ha every boom mus be followed by an equal size bus. Or, more accuraely, if inflaion is going o remain consan in he long run, any negaive unemploymen gap mus evenually be offse by an equal sum of posiive unemploymen gaps laer:
( u u ) [( ) ( u u) 0] 1 T 0 T 0 This implicaion disappears when (1-λ), he coefficien on lagged inflaion is less han one, and a foriori when, as appears o be he case oday, he coefficien is close o zero. In his case, a boom will be associaed wih higher inflaion, bu inflaion will decrease as unemploymen reurns o he naural rae, and here is no need or necessiy for he boom o be followed by a bus. Pu anoher way, here may be no cos o having a emporary boom, excep for emporary higher inflaion. This is where he echo of he policies followed in he 1960s, he painful lessons of he 1970s, and he Lucas criique come in. They raise he quesion of wha exacly lies behind he anchoring of expecaions. I mus be in large par due o moneary policy credibiliy and a long period of low inflaion; in his case, prolonged deviaions of inflaion from arge may deanchor expecaions. Inflaion below arge does no appear o have had his effec so far, bu i is hard o know wha margin moneary policy has before hey do ge de-anchored. Anoher possibiliy is ha he anchoring of expecaions reflecs a lack of salience: A very low raes of inflaion, people may no focus on inflaion, and hus may no adjus expecaions in response o movemens in inflaion. If his is he case, i implies ha he Fed may have some room o use so long as inflaion remains low enough so as o no become salien. The (oo) appealing rade-off beween unemploymen and inflaion. A small slope coefficien θ implies an aracive shor-run radeoff beween inflaion and unemploymen. A value of -0.2 implies ha a 1% decrease in unemploymen for one quarer increases inflaion, measured a an annual rae, by 0.2%. Combine his wih he anchoring of expecaions and a value of λ close o 1, which implies ha, even if unemploymen remains lower,
inflaion will no increase much above 0.2%, and he rade-off becomes even more aracive, raising srong Barro-Gordon empaions o lower unemploymen below he naural rae for some ime. (Hyseresis argumens may provide a valid reason o do so, and his is wha led us o reexplore hyseresis in BCS, and conclude ha hyseresis may indeed well be presen. Bu hey may also provide a smokescreen for succumbing o empaion.) One can already see he pressure on he Fed, for example o no raise raes unil i sees he whies of inflaion s eyes. Given he lags in he effec of higher ineres raes on aciviy, his would appear o be a subopimal policy. The failure of he divine coincidence In he benchmark New-Keynesian model, sabilizing inflaion keeps he unemploymen rae a he naural rae, and he naural rae in urn is he consrained efficien rae, i.e. he bes rae ha can be achieved by policy. Jordi Gali and I have called his proposiion he divine coincidence. Addiional disorions ypically lead o deviaions of he naural rae from he consrained efficien rae, bu he divine coincidence remains a useful heoreical benchmark. The evidence from above is however ha i fails badly empirically: This is refleced by he large sandard deviaion of he residual in he Phillips curve. The residual can be inerpreed in wo ways: Firs as capuring unobserved movemens in he naural rae. If so, i implies large, high frequency, movemens in he naural rae. As he consrained efficien rae is likely o move slowly, his in urn implies large, high frequency, deviaions of he naural rae from he consrained efficien rae. Or i can be inerpreed as he resul of misspecificaion, for example he use of he wrong inflaion series, or he wrong dynamic specificaion. In eiher case, i implies ha he Fed faces a rade-off beween sabilizing unemploymen and sabilizing inflaion. In he language of moneary policy, i needs o go for
very flexible inflaion argeing, wih poenially difficul communicaion problems, especially given he empaions discussed earlier. In shor, he Phillips curve is sill here. Bu is curren shape raises serious challenges for moneary policy in he fuure. References Ball, L., S. Mazumder, 2011, Inflaion Dynamics and he Grea Recession, IMF WP 11/121 Blanchard, O., E. Cerui, L. Summers, 2015, Inflaion and Aciviy. Two Exploraions and Their Moneary Policy Implicaions, in Inflaion and Unemploymen in Europe, ECB Forum on Cenral Banking, www.ecb.europa.eu/pub/pdf/oher/ecbforumoncenralbanking2015en.pdf Inernaional Moneary Fund, 2013, The Dog ha Didn Bark. Has Inflaion Been Muzzled or Was I Jus Sleeping?, World Economic Oulook, Chaper 3, April Kiley, M., 2015, Low Inflaion in he Unied Saes: A Summary of Recen Research, FEDS Noes, November 2015 Krugman, P. Anchors away, 2015, hp://krugman.blogs.nyimes.com/2015/12/04/anchorsaway-slighly-wonkish/