5 PhillipsCurveBasics

P HILLIPS C URVE E STIMATION : BASICS
Emi Nakamura Jón Steinsson
UC Berkeley
September 2021
Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 1 / 68

B RIEF H ISTORY OF THE P HILLIPS C URVE
Phillips 58 points out empirical relationship between wage inflation

and unemployment in UK 1861-1957
Samuelson-Solow 60 popularize idea in US

hypothesis stated above is correct the rise in import prices in 1862 may
just have been sufficient to start up a mild wage-price spiral, but in the
remainder of the period changes in import prices will have had little
I NFLATION AND NEMPLOYMENT IN THEU
or no effect on the rate of change of wage rates.
UK
U 0
E? 2 -
cf.
0 0~~~~
cl)
0 ? -2
. ~~~~~~~~~~*
.-40 1 2 3 4 6 $ 7 8 9 10 11
Unemptoyment, %.
Fig,t. 1861 - 1913
Source: Phillips (1958)
I NFLATION AND U NEMPLOYMENT IN THE UK
286 ECONOMICA [NOVEMBER
10
@i 10
X.
*-,Curve fitted to 1861-1913 data
6-
IV
4 64 63
c 2
E: 2- t-t 2 > ~~~~~62
-2
z ~~~~~~~~~67
I I I * a I I I I I B
0 0 1 2 3 4 5 6 7 8. 9 10 11
Unemployment, %.
Fig.2. 1861 - 1868

Unemployment, %.
Fig.2. 1861 - 1868

2 Curve fitted to 1861-1913 data

72
6-6
73
c)
0
3: 4l
Z'% 70
0 Unmpoyen,74.
Fig.3. 1868 -1879 ~ ~ ~ 7

288 ECONOMICA [NOVEMBER
C
t1:
10-
-f1C Curve fitted to 1861 -1913 data

+' 6 \
06
0)
3 4 \9
o 90
E 87 86
-0
0 92 93
0)C -2
0 1 2 3 4 5 6 7 8 9 10 I
cr. Unemptoyment, %.
Fig.5. 1886- 1893

cr. Unemptoyment, %.
Fig.5. 1886- 1893

># 10
* \ Curve fitted to 1861-1913 data

6-
(U4
no4- \
0 0
E 0
@ ~~~~~01 02 os 04 94 93
X-2-
"o -4J
0 1 2 3 4 5 6 7 8 9 10 11
Unemployment, 1.
Fig.6. 1893 -1904

US M ACRO P OLICY IN THE 1960 S
Phillips curve viewed as a menu of options
Policy makers can lower unemployment if they are

willing to tolerate more inflation

I NFLATION AND U NEMPLOYMENT IN THE US
Inflation
16.0
14.0
12.0
10.0
8.0
6.0 1969
4.0
2.0
1961
0.0
2.0 4.0 6.0 8.0 10.0
Unemployment

F RIEDMAN AND P HELPS M AKE A P REDICTION
Friedman 68 and Phelps 67:

Policymakers cannot exploit a stable Phillips curve forever
Workers will demand wage increases in excess of expected inflation
As inflation rises, expectations of inflation will rise
Changes in expected inflation will shift the Phillips curve

I NFLATION AND U NEMPLOYMENT IN THE US
Inflation
16.0
14.0
12.0
10.0
8.0
6.0 1969
4.0
2.0
1961
0.0
2.0 4.0 6.0 8.0 10.0
Unemployment

F RIEDMAN AND P HELPS W ERE R IGHT !
Inflation
16.0
14.0 1980
12.0
10.0
8.0
6.0 1969
4.0
2.0
1961
0.0
2.0 4.0 6.0 8.0 10.0
Unemployment

M ODERN P HILLIPS C URVE
πt = βEt πt+1 + κ(yt − ytn ) + ηt
Three drivers of inflation:

Expected inflation: Et πt+1
Output relative to potential: yt − ytn
Cost-push shocks: ηt
Specific form above based on Calvo 83 sticky-price assumptions

Details vary across specifications
(e.g., sticky information yields Ēt−1 πt )
Structural equation originating from firm’s price setting decision

E STIMATING S LOPE OF THE P HILLIPS C URVE
Object of interest: Slope coefficient κ

How much does an increase in “demand” / “tightness” / “output gap”
affect inflation

E STIMATING S LOPE OF THE P HILLIPS C URVE
Object of interest: Slope coefficient κ

How much does an increase in “demand” / “tightness” / “output gap”
affect inflation
Tricky identification issues:

Expected inflation unobserved
“Natural rate of output” (i.e., supply shocks) unobserved
Cost push shocks (e.g., variation in desired markups) unobserved
All three may cause omitted variables bias

E XPECTED I NFLATION
Pre Friedman/Phelps Phillips curve: Change in output gap needed

to change inflation
πt = µ + κ(yt − ytn )
Same is true for accelerationist Phillips curve

(i.e., Phillips curve with adaptive expectations)
πt = πt−1 + κ(yt − ytn )

Pre Friedman/Phelps Phillips curve: Change in output gap needed

to change inflation
πt = µ + κ(yt − ytn )
Same is true for accelerationist Phillips curve

(i.e., Phillips curve with adaptive expectations)
πt = πt−1 + κ(yt − ytn )
Sargent 82: Hyperinflations end abruptly with little or no output cost

Clear violation of aforementioned Phillips curves

G ERMAN H YPERINFLATION
Sargent (1982):
The Ends of Four
Big Inflations
9
Source: Sargent (1982)
In Calvo model, perfectly credible, unexpected disinflation can occur

without any effect on output gap
Expected inflation does all the work
Theoretical victory: Potential explanation for Sargent facts

In Calvo model, perfectly credible, unexpected disinflation can occur

without any effect on output gap
Expected inflation does all the work
Theoretical victory: Potential explanation for Sargent facts
Empirical headache:
Movements in inflation potentially completely unrelated to output gap
Even if output gap moves during disinflation, not clear what fraction
of disinflation was due to shift in expected inflation
Measurement of expected inflation crucial but hard

S UPPLY S HOCKS
Estimation of Phillips curve slope also complicated by

classic simultaneity problem
Need to isolate demand variation to estimate slope
Supply shocks yield “stagflation”

(i.e., positive correlation between unemployment and inflation)
Bias slope estimates towards zero (or “wrong” sign)

I NFLATION E XPECTATIONS + S UPPLY S HOCKS
Inflation
16.0
14.0 1980
12.0
10.0
8.0
6.0 1969
1989
4.0
2.0 1983
1961
0.0
2.0 4.0 6.0 8.0 10.0
Unemployment

I S THE P HILLIPS C URVE D EAD ?
Phillips curve often pronounced dead

Many economists think Phillips curve is an empirical disaster
Prominent episodes:
Missing inflation in late 1990s
Missing disinflation in the Great Recession
Missing reinflation in the subsequent recovery
Missing disinflation in the COVID crisis
Seems like inflation is always going missing...

M ISSING I NFLATION IN L ATE 1990 S
Inflation
16.0
14.0
12.0
10.0
8.0
6.0 1990
4.0 2000 1992

2.0
1986
0.0
2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0
Unemployment

M ISSING D ISINFLATION IN THE G REAT R ECESSION
Inflation
16.0
14.0
12.0
10.0
8.0
6.0 1990
4.0 2000 2007

1992
2010
2.0 2003
1986
0.0
2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0
Unemployment

M ISSING R EINFLATION SINCE G REAT R ECESSION
Inflation
16.0
14.0
12.0
10.0
8.0
6.0 1990
4.0 2000 2007

1992
2010
2.0
2017 2014
0.0
2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0
Unemployment

ATKESON AND O HANIAN (2001)
Are Phillips Curves Useful for Forecasting Inflation?

Answer: No
Methodology:
Compare forecasts from Phillips curve models with
“naive” no-change model
Metric of fit: root mean squared error (RMSE)
“Online” estimation using data from January 1959 onward

ATKESON AND O HANIAN (2001)
Naive model:
12
Et πt+12 = πt12
Original Phillips curve:
12
Et πt+12 = β(ut − ū)
NAIRU Phillips curve:
12
Et πt+12 = πt12 + β(ut − ū)
Stock and Watson’s (1999) NAIRU Phillips curve:
12
Et πt+12 = πt12 + α + β(L)ut + γ(L)(πt − πt−1 )
(Their nomenclature)

RMSE VS . “NAIVE ” M ODEL
over the subsequent four

Why Use the NAIRU Phillips Curve? Greenbook.13 Specificall
Ratios of Errors of NAIRU and Naive Model* Forecasts of Inflation for 1984-99, prepared for FOMC m
Made With Alternative Indicators and Measures November 13 for the fou
through 1995. These fo
years 1984 through 1996
Range of Ratio of NAIRU/Naive RMSEs** from the Greenbook in
Inflation Inflation forecasts are confidentia
Indicator Measuret Minimum Maximum in selecting forecasts im
were compiled no more
Unemployment PCE Deflator 1.02 1.34 as much as six weeks la
Rate CPI .99 1.32
torical data used for the
This timing suggests tha
Core CPI 1.06 1.94 be more accurate, on ave
no other reason than that
Activity Index* PCE Deflator 1.04 1.23
when the Greenbook for
We compare both the
CPI 1.06 1.32 inflation forecasts again
Core CPI 1.33 1.81 computed using current d
We follow the design o
ments by comparing the
Stock and Watson (1999)
*The NAIRU NAIRU
models are Phillips
versions of Stockcurve
and Watson' vs. Naive
s (1999a) naive with different forecasts to the RMSE
model
models. The
lag lengths from 1 to 12model simply predicts that at any date inflation will be the same as it had been over
foryear.both β(L) and γ(L). 1984. Wefind that the R
the past
**RMSE=m\ mean squared error.
Nakamura-Steinsson (Berkeley) Phillips Curve
naive forecasts are basic
Sept 2021 26 / 68
The Breakdown in an Early Phillips Curve over the next fo
Quarterly Unemployment as a Percentage of the U.S. Labor Force vs. regression line t
W HY S O HChanges
ARDin the
TO F ORECAST
Implicit Price Deflator for U.S.?GDP
( IN
Over LEVELS )
the Next Four Quarters, preted as a fore
1st Quarter 1959-1 st Quarter 1999 given the curren
is clearly down
Chart 1 A Negative Relationship in 1959-69 . . . negative relation
1960s.4
After 1970, h
vironment chang
er and more vo
1960s. As the e
ative relationshi
tion observed in
1, disappeared.
this negative rel
plays quarterly d
quarter of 1970
inflation rate ov
shows two regr
from Chart 1, co
regression line th
downward-slopi
gression linefro
relationship betw
tion. Moreover,
Nakamura-Steinsson (Berkeley) Phillips Curve based on the 19
Sept 2021 27 / 68
gression linef
W HY S O H ARD TO F ORECAST ? ( IN LEVELS ) relationship be
tion. Moreove
based on the
Chart 2 . . . Disappeared in 1970-1999 curate. Lucas
down of this
Inflation Rate (%)
that of the mo
on it, represen
Thus, both the
not to use Phi
The NAIRU S
Yet some still
that there is a
tween unempl
to forecast inf
versions of th
the early spec
NAIRU Philli
sure of econom
in the inflation
4Brayton et al.
early versions28of/ 68the
Sources:
Nakamura-Steinsson U.S.
(Berkeley) Departments of Labor and Commerce
Phillips Curve Sept 2021
Quarterly Unemployment as a Percentage of the U.S. Labor Force vs.
Difference Between Change in the Implicit Price Deflator for U.S. GDP quarter of 1960
) the vertical axis
Over the Next Four Quarters and Its Change Over the Previous Four Quarters,
W HY S O H 1stARD F
TO st Quarter
Quarter 1960-1 ORECAST1999 ?( IN CHANGES sured by the GD
ters relative to
quarters. The lin
Chart 3 The Steep Negative Relationship in 1960-83 . . . these data. This
cast of the chan
relative to inflat
the current unem
a negative relat
sequent changes
that, during this
was low, there w
and when the u
tendency for th
identifies a NA
rate of unemplo
regression, the i
or fall.
Note that the
book NAIRU P
duced by the lar
Unemployment Rate (% of Labor Force) the Federal Res
that if unemplo
Nakamura-Steinsson (Berkeley) Phillips Curve NAIRU "on a s
Sept 2021 29 / 68
book NAIRU
duced by the la
W HY S O H ARD TO F ORECAST ?Unemployment ( IN CHANGES ) the Federal Res
Rate (% of Labor Force)
that if unemplo
Chart 4 . . . Flattened in 1984-99
NAIRU "on a s
rise about 0.4 o
der, Tetlow, an
that when unem
the NAIRU, at
of a percentage
Of course, t
NAIRU Phillip
instability with
was the early P
to expect the N
many aspects o
1980s: the busi
have all been le
previous 15 ye
6We follow Stock

Unemployment Rate (% of Labor Force) ment rate against subse
Sources: U.S. Departments of Labor and Commerce
NAIRU Phillips curve
7Clarida, Gali, and
Nakamura-Steinsson (Berkeley) Phillips Curve cantly in the early
Sept 2021 198
30 / 68
F ORECASTING AND THE P HILLIPS C URVE
Theory does not suggest that the Phillips curve would necessarily
be useful for forecasting
Phillips curve is a supply curve
Useful for forecasting only if (when) demand variation is dominant

(and inflation expectation stable)
Clearly not true in 1970s and 1980s
Same as any other market

Supply curve for oil not necessarily useful to forecast price of oil

T HREE S TRANDS OF P HILLIPS C URVE L ITERATURE
1. Aggregate Variation with Adaptive or Survey Expectations

Stock-Watson (2010, 2019), Ball-Mazumder (2011, 2019),
Coibion-Gorodnichenko (2015)
2. Aggregate Variation with Rational Expectations

Gali-Gertler (1999), Sbordone (2002),
Mavroeidis-Plagborg-Muller-Stock (2014)
3. Cross-Sectional Variation
Fitzgerald-Nicolini (2014), McLeay-Tenreyro (2019),
Hazell-Herreno-Nakamura-Steinsson (2021)

K EY Q UESTIONS
Has the Phillips curve flattened?
Is there missing disinflation / reinflation?
Does “anchoring” of inflation expectations explain stability of inflation?
Is there a stable Phillips curve?

F LATTENING P HILLIPS C URVE
6
1960-1983
5 1984-1999
Year-over-year change in inflation 2000-2019
4
-1
-2
-3
-4
-3 -2 -1 0 1 2 3 4 5
Unemployment Gap
Figure 1: Stock and Watson’s Changing Phillips Correlation
Note: Black solid line
Nakamura-Steinsson is a regression line for 2000-2019.
(Berkeley) Dark grey broken line is regression for 1984-1999.
Phillips Curve Sept 2021 Light
34 / 68
W HY M IGHT P HILLIPS C URVE H AVE F LATTENED ?
Inflation fell and prices became more sticky

(as menu cost model would predict)
Inflation expectations became better anchored

Output gap and change in inflation expectations correlated
in 1970s and 1980s (biased estimates of Phillips curve slope)
Some other structural change to the economy

I S T HERE A S TABLE P HILLIPS C URVE ?
To “see” the Phillips curve, must control for:

Changes in inflation expectations
Supply shocks
Stock and Watson (2010):

The history of the Phillips curve “is one of apparently stable relationships
falling apart upon publication.”

BALL AND M AZUMDER (2019)
Empirical specification:
πt = πte + α(ut − ut∗ ) + t
Focus on post-1985 period
Use “non-standard series”:

Median inflation
Long-run inflation expectations
Short-term unemployment
Ignore endogeneity

M EDIAN I NFLATION
Basic idea to get away from supply shocks
More common to use core

Supply shocks important in food and energy
Ball and Mankiw (1995):

Relative price changes (due to supply shocks) can affect aggregate
inflation in a menu cost model
Firms in sectors with large shocks will adjust, while others will not
Ball and Mazumder (2011, 2019): median inflation filters out

movements in headline inflation due to large relative price movements
in all sectors (not just food and energy)

M:EDIAN
118 I NFLATION
MONEY, CREDIT AND BANKING
FIG.
Nakamura-Steinsson (Berkeley) 3. Median CPI and CPIX
PhillipsQuarterly
Curve Inflation, 2000–15. Sept 2021 39 / 68
S HORT-RUN U NEMPLOYMENT
Literature uses various different “slack” measures
Rationale for short-run unemployment:

Long-term unemployed are on the margins of the labor force
Don’t put pressure on wages
Largely co-linear with total unemployment prior to Great Recession
Not so during Great Recession

(smaller rise results in smaller fitted fall in inflation)

S HORT -RUNCREDIT
116 : MONEY, U NEMPLOYMENT
AND BANKING R ATE
FIG. 2.
Nakamura-Steinsson (Berkeley) Short-Term Unemployment versus
Phillips Total Unemployment, 1985–2015.
Curve Sept 2021 41 / 68
L ONG -RUN I NFLATION E XPECTATIONS ?
Which inflation expectations should be used?
Ball and Mazumder (2019) use long-run SPF inflation forecasts
Doesn’t New Keynesian model say one should use one-period-ahead

inflation expectations?
Can one just pick whatever one want’s?
We will come back to this (when discussing cross-sectional papers)

L ONG -RUN I NFLATION E XPECTATIONS
LAURENCE BALL AND SANDEEP MAZUMDER : 115
FIG. 1. Long-Term
Nakamura-Steinsson SPF Inflation
(Berkeley) Expectations versusPhillips
Four-Quarter
Curve Moving Average of Median Inflation, 1985–2015.
Sept 2021 43 / 68
P HILLIPS C URVE E STIMATION
120 : MONEY, CREDIT AND BANKING
TABLE 1
AN EXPECTATIONS-AUGMENTED PHILLIPS CURVE, 1985–2015
πt = πte + α(u st−1 − u s,∗

t−1 ) + t
α −0.756
(0.077)
DW 1.259
SE of Reg. 0.383
2
R 0.824
NOTE: OLS with Newey–West (1987) standard errors in parentheses. πt is median CPI inflation, πte is the average forecast of long-term CPI
inflation from the Survey of Professional Forecasters, u st−1 is the average of the short-term unemployment rate from t − 1 to t − 4, and u s∗
t−1
is the average of the natural rate of short-term unemployment from t − 1 to t − 4.
2.3 Results
Table 1 presents estimates of the Phillips curve in equation (5) for our 1985–2015
sample. The estimated
Nakamura-Steinsson (Berkeley) coefficient onPhillips
short-term
Curve unemployment is −0.76,
Sept which
2021 44 / 68
G OOD F IT ! LAURENCE BALL AND SANDEEP MAZUMDER : 121
(a)
π
(b)
G OOD F IT !
π
(b)
π − πe

S TABLE ACROSS S UB - PERIODS
122 : MONEY, CREDIT AND BANKING
TABLE 2
STABILITY OF THE EXPECTATIONS-AUGMENTED PHILLIPS CURVE
πt = πte + α(u st−1 − u s,∗

t−1 ) + t
1985Q1–1997Q4 1998Q1–2007Q4 2008Q1–2015Q4
α −0.702 −0.781 −0.795

(0.094) (0.228) (0.109)
DW 1.492 1.043 1.286
SE of Reg. 0.361 0.436 0.353
2
R 0.764 0.316 0.755
p-Value for stability 0.813
NOTE: OLS with Newey–West (1987) standard errors in parentheses. πt is median CPI inflation, πte is the average forecast of long-term CPI
inflation from the Survey of Professional Forecasters, u st−1 is the average of the short-term unemployment rate from t − 1 to t − 4, and u s∗
t−1
is the average of the natural rate of short-term unemployment from t − 1 to t − 4. The reported p-value is for a Wald test of the hypothesis
that α is equal in the three subsamples.
The results, reported in Table 3, yield two clear conclusions. First, for the entire
sample, the data
Nakamura-Steinsson say that short-term unemployment
(Berkeley) Phillips Curve is the right variable in theSept
Phillips
2021 47 / 68
H AVE E XPECTATIONS B ECOME M ORE A NCHORED ?
Anchored Expectations:
πte = 2.5 + t
Backward-Looking Expectations:
1
πte = [(1 − γ)πt−1 + γ(1 − γ)πt−2 + ... + γ 39 (1 − γ)πt−40 ] + t
1 − γ 40
Nested Specification:
Weighted average with weight λ
Ball-Mazumder estimate this allowing for a break in λ

Estimated break date is 1998Q1

A NCHORING OF E XPECTATIONS
LAURENCE BALL AND SANDEEP MAZUMDER : 125
TABLE 4
ANCHORED VS. BACKWARD-LOOKING EXPECTATIONS
πte = λ2.5 + (1 − λ) 1 [(1 − γ )πt−1 + γ (1 − γ )πt−2 + . . . + γ 39 (1 − γ )πt−40 ] + t

1−γ 40
1985Q1–2015Q4 (with 1998Q1 Break in λ)
λ pr ebr eak
0.067 0
(0.046)
λ postbr eak 0.773 1
(0.066)
γ 0.875 0.859
(0.018) (0.017)
DW 0.357 0.312
S E of Reg. 0.189 0.203
2
R 0.940 0.930
NOTE: NLLS with Newey–West (1987) standard errors in parentheses. πte is the average forecast of long-term CPI inflation from the Survey
of Professional Forecasters, and πt is median CPI inflation. The break date of 1998Q1 is the quarter that produces the largest Wald statistic
for the hypothesis that λ pr ebr eak =λ postbr eak .

A NCHORING
126: MONEY, CREDIT AND E XPECTATIONS
OF BANKING
Postbreak 1,

C OUNTERFACTUAL WITH NLAURENCE
O A NCHORING
BALL AND SANDEEP MAZUMDER : 127
1985—97 1998—2015

C OIBION -G ORODNICHENKO 15
Focus on “missing disinflation” during Great Recession
Argue that population explanations insufficient

Anchored inflationary expectations
Movements in natural rate
Flattening of the Phillips curve
New explanation:
Household inflation expectations rose in 2009-2013
If firm’s expectation the same, this can explain missing disinflation

Baseline assumptions:
Output gap measure: Unemployment rate
yt − ytn = ut
(Ignore natural rate utn )
Expectations of inflation: backward looking
1
Et πt+1 = (πt−1 + πt−2 + πt−3 + πt−4 )
4
(Ignore discounting: β = 1)

Back = κu + η
πt − Eπt+1 t t
Estimate by OLS for sample 1960Q1-2007Q4

Implicitly assuming that ηt ⊥ ut (i.e., ignoring supply shocks)
Consider alternative specifications later
See whether Great Recession “sticks out”

M202
ISSING D ISINFLATION : CPI
AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2015
Panel A. CPI inflation and US unemployment
09Q4
09Q3
4
11Q1
07Q4 11Q2
2 09Q2
08Q3 10Q4
08Q2
08Q1 12Q4
0 12Q3
13Q1 11Q3 10Q3
πt−EπtBack
10Q1
12Q1
−2 12Q2
11Q4
10Q2
−4 09Q1
−6
−8
3 4 5 6 7 8 9 10 11
Unemployment rate
Source: Coibion and Gorodnichenko (2015)
3 4 5 6 7 8 9 10 11
M ISSING D ISINFLATIONUnemployment rate
Panel B. CPI inflation and predicted inflation from the Phillips Curve
8
2
Inflation
−2
−4
−6
Actual
−8 Backward PC
−10
2007 2008 2009 2010 2011 2012 2013
Figure 1. The Missing Disinflation

Notes: Panel A shows
Nakamura-Steinsson the scatter plot of inflation surprises
(Berkeley) (π t − E tπ BACK
Phillips Curve t  ) versus unemployment rate. 2021
SeptE  tπ BACK
t  56
is / 68
08Q3 12Q3 11Q3 10Q309Q4
0 13Q1
12Q3
09Q210Q4
0
08Q2 11Q2
PCE inflat
12Q4 12Q1 10Q4
πt − EπtBA
11Q1
πt − EπtB
08Q1 10Q1 12Q1
GDP deflator i
11Q3
08Q1
10Q3 07Q4 12Q412Q2
13Q1 09Q3
−2 −2
10Q2 09Q1
11Q4
09Q1
12Q2
M ISSING
−4
D ISINFLATION : C ORE CPI/PCE
−4
11Q4
09Q2
−6 −6
−8 −8
3 4 5 6 7 8 9 10 11 3 4 5 6 7 8 9 10 11
Unemployment rate Unemployment rate
Panel C. Core CPI inflation Panel D. Core PCE inflation
4
4
PCE CORE inflation

CPI CORE inflation
2
2 11Q2 09Q4
πt − EπtBACK
πt − EπtBACK
07Q4 12Q1 11Q3
11Q2
11Q1
11Q3 08Q2
07Q4 09Q4 0 08Q1
13Q1 12Q2
09Q3
11Q109Q2
10Q2
10Q1
0
08Q1 08Q3 13Q1 12Q2 09Q2 08Q3 12Q4
12Q3 11Q4
11Q4
12Q1 10Q3
10Q4 10Q4
10Q3
12Q4 09Q1
−2
08Q2 12Q3 09Q3
10Q2 09Q1
−2
08Q4
10Q1
−4 −4
−6 −6
−8 −8
3 4 5 6 7 8 9 10 11 3 4 5 6 7 8 9 10 11
PanelCoibion
Source: E. SPF and
inflation (CPI) forecasts
Gorodnichenko (2015) Panel F. Controlling for oil prices
4
e changes
09Q4
08Q3
4 09Q3
07Q4 08Q2 11Q1
11Q2 11Q1
2 08Q1
09Q3 2
08Q3
11Q2
09Q4
10Q4
SPF
10Q4
ACK
11Q3 07Q4 12Q4

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021
11Q3 57 / 68
t
A LTERNATIVE SPECIFICATIONS
Survey expectations
CBO estimates of natural rate
Oil shocks
Help address alternative explanations:

Anchoring of inflation expectations
Movements in natural rate
Role of supply shocks

−2
πt −
πt −
−4
PCE C
−4
CPI C
M ISSING
−6 −6
D ISINFLATION : SPF F ORECAST /O IL C ONTROL
−8 −8
3 4 5 6 7 8 9 10 11 3 4 5 6 7 8 9 10 11
Panel E. SPF inflation (CPI) forecasts Panel F. Controlling for oil prices
partial out oil price changes

09Q4
08Q3
4 09Q3
07Q4 08Q2 11Q1
11Q2 11Q1
2 08Q1
09Q3 2
08Q3
11Q2
09Q4
10Q4
πt − EπtSPF
10Q4
πt − EπtBACK
11Q3 07Q4 12Q4
11Q3
12Q1 09Q2 08Q1 12Q3
0 12Q4
12Q3
10Q3 0 08Q2
09Q2
10Q1
13Q1 11Q4 10Q3
10Q1 13Q1
12Q2 12Q2
12Q1
−2 −2
09Q1
10Q2 10Q2
11Q4
−4 09Q1
−4
−6 −6
3 4 5 6 7 8 9 10 11 −2 0 2 4 6
Unemployment rate,
Unemployment rate partial out oil price changes
Figure 2. Robustness of the Missing Disinflation

Source: Coibion and Gorodnichenko (2015) – SPF forecast over next four quarters.
Note: See notes for Figure 1 and the text for more details.

M ISSING
VOL.7 NO.1
D ISINFLATION : NATURAL R ATE
COIBION AND GORODNICHENKO: THE MISSING DISINFLATION 207
Panel A. Missing disinflation with CBO unemployment gaps
09Q4
09Q3
4
11Q1
07Q4 11Q2
2 08Q3
09Q2
10Q4
08Q2
08Q1 12Q4
0 12Q3
πt − EπtBACK
13Q1 11Q3 10Q3

10Q1
12Q1
−2 12Q2
11Q4
10Q2
−4 09Q1
−6
−8
−2 −1 0 1 2 3 4 5
Unemployment gap
−2 −1 0 1 2 3 4 5
M ISSING D ISINFLATION : Unemployment

NATURALgapR ATE
Panel B. Changes in natural rate of unemployment needed to explain missing disinflation
12
10
4 Predicted natural rate

Actual CBO natural rate
2 95% CI for predicted rate

Actual unemployment rate
2007 2008 2009 2010 2011 2012 2013

Figure 3. Does the Missing Disinflation Reflect Structural Unemployment?
N EW E XPLANATION
Expectations typically measured by SPF forecasts
But is this the way to go?
Perhaps firm expectations exhibit similar biases to household

expectations
Show that household expectations (Michigan survey) have quite

different properties from SPF
Overreact to gasoline prices

H OUSEHOLD
218
E XPECTATIONS VS . SPF F ORECASTS
AMERICAN ECONOMIC JOURNAL: MACRO
Panel A. Inflation expectations Panel B. P

for different economic agent inflation e
Asset prices
4
Michigan
12
SPF (CPI) 2
10
πt − EπtMSC
8 0
6
−2
4
−4
2
0 −6
1980 1985 1990 1995 2000 2005 2010 −3 −

I NFLATION E XPECTATIONS
ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2015
Panel B. Phillips Curve with household

inflation expectations
Asset prices
4 1960Q1−1984Q4
Michigan 1985Q1−2007Q3
SPF (CPI) 2 2007Q3−2013Q1
07Q4 08Q3
πt − EπtMSC
08Q1 09Q3
0 11Q1 09Q4
11Q2 10Q4
08Q2
11Q3 09Q2
−2 12Q 4QQ31
12
12 10Q3
11Q4
13Q12
1 Q2 10Q1
−4 10Q2
09Q1
−6
2005 2010 −3 −2 −1 0 1 2 3 4 5
Unemployment gap
Panel D. Relative contribution

Nakamura-Steinsson (Berkeley) Phillips Curve
of slopes Sept 2021 64 / 68
I NFLATION E XPECTATIONS
Three differences versus SPF:

No evidence of flattening
Flatter throughout
No evidence of missing disinflation!

S TOCK AND WATSON (2019)
Large variation across sectors in correlation between inflation and

cyclical component of real activity
Stronger correlation for well-measured, domestic components
In particular housing
Median inflation measure used by Ball-Mazumder 19 ends up

placing a lot of weigh on housing

M EASUREMENT OF H OUSING
Inflation measure for owner-occupied housing changed in 1983

pre-83: Changes in house prices and mortgage costs (interest rates)
post-83: Changes in rents
Makes a BIG difference for properties of CPI
CPI Research Series uses modern methodology back in time

(as do PCE and GDP deflators)

W HAT IF W E U SE O LD M ETHOD FOR R ECENT P ERIOD ?
Pre-1983 Methods
Post-1983 Methods
15
10
-5
1975 1980 1985 1990 1995 2000 2005 2010 2015
Date
Figure B.1: CPI Inflation Using Pre- and Post-1983 Housing Methodology
Note: This figure
Nakamura-Steinsson plots overall CPI inflation in the
(Berkeley) US Curve
Phillips (gray line) and our attempt at estimating whatSept
CPI2021 68 / 68

5 PhillipsCurveBasics

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

5 PhillipsCurveBasics

Uploaded by

Copyright:

Available Formats

P HILLIPS C URVE E STIMATION : BASICS

Emi Nakamura Jón Steinsson

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 1 / 68

Phillips 58 points out empirical relationship between wage inflation

Samuelson-Solow 60 popularize idea in US

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 2 / 68

*-,Curve fitted to 1861-1913 data

Fig.2. 1861 - 1868

Fig.2. 1861 - 1868

2 Curve fitted to 1861-1913 data

Fig.3. 1868 -1879 ~ ~ ~ 7

Source: Phillips (1958)

-f1C Curve fitted to 1861 -1913 data

Fig.5. 1886- 1893

Source: Phillips (1958)

Fig.5. 1886- 1893

* \ Curve fitted to 1861-1913 data

Fig.6. 1893 -1904

Phillips curve viewed as a menu of options

Policy makers can lower unemployment if they are

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 8 / 68

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 9 / 68

Friedman 68 and Phelps 67:

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 10 / 68

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 11 / 68

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 12 / 68

πt = βEt πt+1 + κ(yt − ytn ) + ηt

Three drivers of inflation:

Specific form above based on Calvo 83 sticky-price assumptions

Structural equation originating from firm’s price setting decision

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 13 / 68

πt = βEt πt+1 + κ(yt − ytn ) + ηt

Object of interest: Slope coefficient κ

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 14 / 68

πt = βEt πt+1 + κ(yt − ytn ) + ηt

Object of interest: Slope coefficient κ

Tricky identification issues:

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 14 / 68

Pre Friedman/Phelps Phillips curve: Change in output gap needed

Same is true for accelerationist Phillips curve

πt = πt−1 + κ(yt − ytn )

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 15 / 68

Pre Friedman/Phelps Phillips curve: Change in output gap needed

Same is true for accelerationist Phillips curve

πt = πt−1 + κ(yt − ytn )

Sargent 82: Hyperinflations end abruptly with little or no output cost

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 15 / 68

In Calvo model, perfectly credible, unexpected disinflation can occur

Theoretical victory: Potential explanation for Sargent facts

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 17 / 68

In Calvo model, perfectly credible, unexpected disinflation can occur

Theoretical victory: Potential explanation for Sargent facts

Measurement of expected inflation crucial but hard

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 17 / 68

Estimation of Phillips curve slope also complicated by

Need to isolate demand variation to estimate slope

Supply shocks yield “stagflation”

Bias slope estimates towards zero (or “wrong” sign)

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 18 / 68

Nakamura-Steinsson (Berkeley) Phillips Curve Sept 2021 19 / 68

Phillips curve often pronounced dead

Seems like inflation is always going missing...

πt = πte + α(ut − ut∗ ) + t