discussion “in search of a nominal anchor: what drives
TRANSCRIPT
DISCUSSIONldquoIn search of a nominal anchorWhat drives long-term inflation
expectationsrdquoby Carvalho EusepiMoench and Preston
Elmar Mertens
Federal Reserve Board
The results presented here do not necessarily representthe views of the Federal Reserve Systemor the Federal Open Market Committee
September 29 2016
OVERVIEW
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
OVERVIEW
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 20160
05
1
15
2
25
3
35
4
45
5
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
OVERVIEW
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
OVERVIEW
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 20160
05
1
15
2
25
3
35
4
45
5
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
OVERVIEW
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 20160
05
1
15
2
25
3
35
4
45
5
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 20160
05
1
15
2
25
3
35
4
45
5
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 20160
05
1
15
2
25
3
35
4
45
5
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 20160
05
1
15
2
25
3
35
4
45
5
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
ANCHORED INFLATION EXPECTATIONSChairman Bernanke Speech at the NBER SI July 10 2007
bull I use the term rdquoanchoredrdquo to mean [long-runexpectations that are] relatively insensitive to incomingdata
bull the extent to which they are anchored can changedepending on economic developments and (mostimportant) the current and past conduct of monetarypolicy
bull If inflation [runs] higher than [expected] but expectation[s] change little then inflationexpectations are well anchored
bull If the public reacts to higher-than-expectedinflation by marking up their long-run expectationconsiderably then expectations are poorly anchored
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SPF LONG-RUN EXPECTATIONS PCE
Average PCE inflation over next 10 years mean response
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
REALIZED INFLATION AND SPF PCE
PCE headline inflation (monthly change APR) SPF 10-year expectation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue 12m-inflation
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
REALIZED INFLATION AND SPF PCE
black inflation red SPF-10Y blue EWMA
2008 2009 2010 2011 2012 2013 2014 2015 2016
minus10
minus5
0
5
10
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus K)τtminus1|tminus1 + Kπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UC model
πt = τt + πt τt = τtminus1 + ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
TREND INFLATION 101
EWMA Trend is your Friend
τt|t = (1 minus Kt)τtminus1|tminus1 + Ktπt
bull simple filter for persistent component
bull Muth (1961 ECA) Optimal filter in ldquolocal level modelrdquo
Local level UCSV model
πt = τt + πt τt = τtminus1 + σηt ηt πt sim mds
econometricianrsquos BN-trend τt|t = E(πt+infin|πt )
UC BN-trend τt = E(πt+infin|τ t )
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
TIME-VARYING EWMA WEIGHT ldquoUCSVrdquoStock and Watson (2006 JMCB) partπt+infinpartet = Kt
1965 1970 1975 1980 1985 1990 1995 2000 2005 20100
01
02
03
04
05
06
07
08
09
1
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12SRVINFTRM
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
MULTIVARIATE TREND ESTIMATES CPI
τt|(trimmed CPI etc) τt|(SPF etc) from Mertens (REStat forth)
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 20150
2
4
6
8
10
12
SRVINFTRMCEMP
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
THIS PAPER
Ingredients
bull DSGE model with NK-Phillips Curve
bull Twist Learning with ldquoSS bandsrdquo for updating
πt =9830541 minus kminus1
tminus1
983055πtminus1 + kminus1
tminus1
983070πt minus (Etminus1πt minus πtminus1)
983071
kt =
983094ktminus1 + 1 if |Φ(past FE)| lt ν
gminus1 otherwise
Endogenous data feeds into learning behavior
bull Endogenous trend
bull State-dependent sensitivity of trend πt to data
bull Once anchored can tolerate deviations of inflation fromtrend up to a point
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
AGENDA
1 The Problem (aka the motivation)
2 The Paper (aka the solution)
3 The Praise (aka my thoughts)
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2
True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SKETCH OF AN ALTERNATIVE MODEL
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τ
t
+ γπtminus1)
τt=τtminus1 + ηt
Monetary policy with inflation target τ
t
it = rt + τ
t
+ φπ(πt minus τ
t
) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB)
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt = τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SKETCH OF AN ALTERNATIVE MODELelements from Ireland (2000 JMCB) Erceg amp Levin (2003 JME) etc
NK Phillips Curve
πt = βπt+1|t + κxt + ut
πt = πt minus ((1 minus γ)τt + γπtminus1) τt=τtminus1 + ηt
Monetary policy with inflation target τt
it = rt + τt + φπ(πt minus τt) + φxxt + εt
Public has limited information
πt+1|t = E(πt+1|Zt Zt =983056πt xt it
983057)
τt|t = τtminus1|tminus1 + K(Zt minus Zt|tminus1)
Simulate with K ∕= 0 while Vol (ηt) = 0
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3
State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4
Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper
SOME THOUGHTS
1 ldquoExogenousrdquo vs ldquoendogenousrdquo trend
bull π = Etπt+infin vs πt = Etπt+infinbull τt vs τt|t
2 True end-point of inflation
bull Does DGP satisfy 1T
983134t πt rarr E(πt) = π
bull What is your estimate
3 State-dependent gainWhich sizekind of structural shock (and policy response)makes updating behavior switch
4 Fitting long-term forecastsSuppose survey forecasts and inflation are cointegratedrsquolsquoCloserdquo fit better than mimicking same low-frequencybehavior
CONCLUSION
Key question
What drives long-run inflation expectations
This paper
bull State-dependent sensitivity of πt to incoming data
bull Learning with ldquoSS bandsrdquofor updating behavior
bull Induces endogenous variations in trend inflation
Great Paper