View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Online Appendix
The Usefulness of the Median CPI in Bayesian VARs used for
Macroeconomic Forecasting and Policy

Brent Meyer and Saeed Zaman

February 2013
Revised: October 2016

This document is a companion to the paper that has the same title as this one.

*Brent Meyer is an assistant policy adviser and an economist at the Federal Reserve Bank of Atlanta
(brent.meyer@atl.frb.org). Saeed Zaman is an economist at the Federal Reserve Bank of Cleveland
(saeed.zaman@clev.frb.org). The views expressed herein are solely those of the authors and do not
necessarily reflect the views of the Federal Reserve Bank of Atlanta, Federal Reserve Bank of Cleveland
or the Federal Reserve System. Brent Meyer thankfully acknowledges that much of the research
documented in this paper was performed while he was a member of the Research Department at the
Cleveland Fed. The authors thank Todd Clark, Domenico Giannone, Edward Knotek II, Ellis Tallman,
and Randy Verbrugge for their helpful criticisms and guidance. They also thank conference participants
at the 35th International Symposium on Forecasting in Riverside, California.

SECTION A1
Name

BVAR1

Name

BVAR2

Name

A Description of the BVAR models we employ

Literature
Referenced

Beauchemin
and Zaman
(2011)

Literature
Referenced
Banbura,
Giannone, and
Reichlin (2010)

Literature
Referenced
Banbura,
Giannone, and
Reichlin (2010)

BVAR3

Christiano,
Eichenbaum,
and Evans
(1999)

Variables (transformations)

Hyperparameters

Frequency

Real GDP (log-level)
Unemployment rate (level)
Consumer Price Index (log-level)
Consumer Price Index ex food and energy (loglevel)
Effective federal funds rate (level)
Nonfarm business compensation (log-level)
Nonfarm business productivity (log-level)
Real personal consumption expenditures (log-level)
Real personal disposable income (log-level)
Payroll employment: total nonfarm (log-level)
KR-CRB spot commodity price index: all
commodities (log-level)
10-year Treasury note yield at constant maturity
(level)
Moody’s seasoned Aaa corporate bond yield (level)
S&P 500 composite stock price index (log-level)
S&P 500 composite dividend yield (level)
Trade weighted exchange value of the US$ vs.
major currencies (log-level)

λ = max(ML)
µ = max(ML)

Quarterly

Variables (transformations)

Hyperparameters

Frequency

Payroll employment: total nonfarm (log-level)
Consumer Price Index (log-level)
Consumer Price Index ex food and energy (loglevel)
Effective federal funds rate (level)

λ=∞
µ=∞

Monthly

Variables (transformations)

Hyperparameters

Frequency

Payroll employment: total nonfarm (log-level)
Consumer Price Index (log-level)
Consumer Price Index ex food and energy (loglevel)
Effective federal funds rate (level)
Index of sensitive materials prices (log-level)
Money stock: M2 (log-level)
Depository Institutions Reserves: Total (log-level)
Depository Institutions Reserves: Nonborrowed
(log-level)

λ = 0.262
µ = 10*0.262

Monthly

Name

Literature
Referenced

Banbura,
Giannone, and
Reichlin (2010)
BVAR4

Name

BVAR5

Christiano,
Eichenbaum,
and Evans
(1999)

Literature
Referenced

Carriero,
Clark, and
Marcellino
(2015)

Variables (transformations)

Hyperparameters

Frequency

Payroll employment: total nonfarm (log-level)
Consumer Price Index (log-level)
Consumer Price Index ex food and energy (loglevel)
Effective federal funds rate (level)
Unemployment rate (level)
Personal income less transfer payments (log-level)
Manufacturing capacity utilization (level)
Industrial production (log-level)
Housing starts (log-level)
Producer Price Index: finished goods (log-level)
Average hourly earnings (log-level)
M1 (log-level)
S&P 500 composite stock price index (log-level)
Index of sensitive materials prices (log-level)
Money stock: M2 (log-level)
Depository Institutions Reserves: Total (log-level)
Depository Institutions Reserves: Nonborrowed
(log-level)
10-year Treasury note yield at constant maturity
(level)
Trade weighted exchange value of the US$ vs.
major currencies (log-level)

λ = 0.108
µ = 10*0.108

Monthly

Variables (transformations)

Hyperparameters

Frequency

Unemployment rate (level)
Consumer Price Index (log-change, annualized
rate)
Consumer Price Index ex food and energy (logchange, annualized rate)
Payroll employment: total nonfarm (log change,
annualized rate)
Weekly hours worked (level)
Initial claims for unemployment insurance (level)
Nominal retail sales (log change, annualized rate)
UM Index of Consumer Sentiment (level)
Single-family housing starts (log change)
Industrial production (log change, annualized rate)
Manufacturing capacity utilization (level)
ISM PMI: Index of supplier delivery times (level)
ISM PMI: Index of new orders (level)
West Texas Intermediate spot price (log change)
Effective federal funds rate (level)
S&P 500 (log change)
10-year Treasury note yield at constant maturity
(level)
Trade weighted exchange value of the US$ vs.
major currencies (log-level)

λ = 0.2
µ=1
τ=1

Monthly

Name

BVAR6

Name

BVAR7

Literature
Referenced

Carriero,
Clark, and
Marcellino
(2015)

Literature
Referenced

Beauchemin
and Zaman
(2011)

Variables (transformations)

Hyperparameters

Frequency

Unemployment rate (level)
PCE Price Index (log-change, annualized
rate)
PCE Price Index ex food and energy (logchange, annualized rate)
Payroll employment: total nonfarm (log change,
annualized rate)
Weekly hours worked (level)
Initial claims for unemployment insurance (level)
Nominal retail sales (log change, annualized rate)
UM Index of Consumer Sentiment (level)
Single-family housing starts (log change)
Industrial production (log change, annualized rate)
Manufacturing capacity utilization (level)
ISM PMI: Index of supplier delivery times (level)
ISM PMI: Index of new orders (level)
West Texas Intermediate spot price (log change)
Effective federal funds rate (level)
S&P 500 (log change)
10-year Treasury note yield at constant maturity
(level)
Trade weighted exchange value of the US$ vs.
major currencies (log-level)

λ = 0.2
µ=1
τ=1

Monthly

Variables (transformations)

Hyperparameters

Frequency

Real GDP (log-level)
Unemployment rate (level)
PCE Price Index (log-level)
PCE Price Index ex food and energy (log-level)
Effective federal funds rate (level)
Nonfarm business compensation (log-level)
Nonfarm business productivity (log-level)
Real personal consumption expenditures (log-level)
Real personal disposable income (log-level)
Payroll employment: total nonfarm (log-level)
KR-CRB spot commodity price index: all
commodities (log-level)
10-year Treasury note yield at constant maturity
(level)
Moody’s seasoned Aaa corporate bond yield (level)
S&P 500 composite stock price index (log-level)
S&P 500 composite dividend yield (level)
Trade weighted exchange value of the US$ vs.
major currencies (log-level)

λ = max(ML)
µ = max(ML)

Quarterly

Table A2: Forecast Comparison of BVAR in Beauchemin and Zaman (2011) --- Pre-Crisis
Relative Mean Squared Error (RMSE) --- relative to BVAR1
A2a: BVAR1 vs. BVAR1 with median CPI inclusion

Real GDP growth

h=1Q
1.002

h=2Q
0.999

h=3Q
1.011

h=4Q
1.007

h=5Q
1.002

h=6Q
1.011

h=7Q
1.011

h=8Q
1.005

Core CPI inflation

0.946

0.946

0.931

0.938

0.935

0.933

0.945

0.953

Headline CPI inflation

0.968

0.968

0.967

0.941

0.940

0.950

0.949*

0.950**

Unemployment Rate

1.007

1.006

1.012

1.016

1.017

1.018

1.018

1.019

Fed Funds Rate

0.999

0.985

0.975

0.967

0.962

0.960*

0.956*

0.958

A2b: BVAR1 vs. BVAR1 with median CPI replacing the core CPI

Real GDP growth

h=1Q
1.023

h=2Q
1.011

h=3Q
1.006

h=4Q
1.005

h=5Q
0.996

h=6Q
1.007

h=7Q
1.006

h=8Q
0.996

Core CPI inflation
Headline CPI inflation

----0.966

----0.960

----0.964

----0.938

----0.938

----0.954

----0.959

----0.966

Unemployment Rate

1.018

1.011

1.030

1.027

1.022

1.018

1.014

1.011

Fed Funds Rate

0.986

0.978

0.974

0.966

0.958

0.955

0.947*

0.946*

Notes for the table: The table A2a lists the mean squared forecast error (MSFE) of the modified Bayesian
VAR with Median CPI added to it relative to the mean squared forecast error of the modified BVAR1.
The table 1b lists the mean squared forecast error (MSFE) of the modified BVAR1 in which core CPI is
replaced with Median CPI relative to the mean squared forecast error of the modified BVAR1. The
reported RMSFEs are for the real GDP growth (quarterly at annual rate), core CPI inflation (quarterly at
annual rate), headline CPI inflation (quarterly at annual rate), the unemployment rate, and the federal
funds rate for h=1,2,…8 step head forecasts (i.e. 2 years out) for the evaluation period 1987Q1 –
2007Q3. Numbers in the bold indicate that the MSFE of the modified BVAR1 with Median CPI is less than
the MSFE from the modified BVAR1.
*denotes significance at 5% level
**denotes significance at 10% level
(based on modified Diebold-Mariano test)

Table A3: Forecast Comparison of BVARs in Banbura et al. 2010 --- Pre-Crisis
Relative Mean Squared Error --- relative to BVARs in Banbura at al 2010
SMALL BVAR (BVAR2) in Banbura et al. 2010

Payroll growth

A3a: BVAR2 vs. BVAR2 with median CPI inclusion
h=1M h=6M h=9M
h=12M
h=15M
h=18M
1.036 0.968
0.984
0.966
0.973
1.065

h=21M
1.126

h=24M
1.187

Headline CPI inflation

1.008

1.048

0.969

0.943

0.967

0.901

0.881

0.878

Fed Funds Rate

0.961

1.043

1.068

1.074

0.997

0.884

0.819

0.793

A3b: BVAR2 with Core CPI vs. BVAR2 with median CPI
h=1M h=6M h=9M
h=12M
h=15M
h=18M
1.014 1.072
1.066
0.943
0.854
0.808

h=21M
0.805

h=24M
0.840

Payroll growth
Headline CPI inflation

0.958

0.929

0.924

0.920

0.940

0.925

0.962

0.982

Fed Funds Rate

0.922

0.836

0.898

1.039

1.049

1.018

0.974

0.947

A3c: BVAR3 vs. BVAR3 with median CPI inclusion
h=1M h=6M h=9M
h=12M
h=15M
h=18M
0.980 1.000
1.024
1.015
1.023
1.039

h=21M
1.061

h=24M
1.096

CEE BVAR (BVAR3) in Banbura et al. 2010

Payroll growth
Headline CPI inflation

1.026

1.026

1.016

0.962

0.995

0.978

0.960

0.967

Fed Funds Rate

0.940

0.994

1.017

0.990

0.972

0.954

0.922

0.901

A3d: BVAR3 with Core CPI vs. BVAR3 with median CPI
h=1M h=6M h=9M
h=12M
h=15M
h=18M
1.031 1.060
1.048
1.033
1.026
1.035

h=21M
1.072

h=24M
1.093

Payroll growth
Headline CPI inflation

1.000

1.011

1.014

0.969

1.007

0.998

0.986

0.993

Fed Funds Rate

0.977

1.037

1.053

1.058

1.042

1.022

0.985

0.957

Medium BVAR (BVAR4) in Banbura et al. 2010

Payroll growth

A3e: BVAR4 vs. BVAR4 with median CPI inclusion
h=1M h=6M h=9M
h=12M
h=15M
h=18M
0.990 1.001
1.006
1.006
1.005
1.003

h=21M
1.008

h=24M
1.018

Headline CPI inflation

0.997

0.990

0.996

0.960**

0.982

0.979*

0.973**

0.982

UR

0.999

0.986

0.995

1.004

1.010

1.013

1.014

1.015

Fed Funds Rate

0.960

0.952*

0.970

0.968

0.967

0.962

0.947

0.933*

A3f: BVAR4 with Core CPI vs. BVAR4 with median CPI
h=1M h=6M h=9M
h=12M
h=15M
h=18M
1.008 1.013
1.004
1.003
1.008
1.014

h=21M
1.042

h=24M
1.045

Payroll growth
Headline CPI inflation

0.996

0.991

0.998

0.960**

0.981

0.982

0.973*

0.984

UR

1.005

1.030

1.054

1.052

1.046

1.039

1.031

1.027

Fed Funds Rate

1.006

0.994

0.988

0.981

0.975

0.969

0.951

0.933

Notes for the table: The tables A3a-A3f lists the mean squared forecast error (MSFE) of the BVAR with
Median CPI added to it relative to the mean squared forecast error of the BVAR without it. The reported
RMSFEs are for the payroll growth (monthly at annual rate), core CPI inflation (monthly at annual rate),
headline CPI inflation (monthly at annual rate), the unemployment rate, and the federal funds rate for
h=1,2,…24 step head forecasts (i.e. 2 years out) for the evaluation period 1987M1 – 2007M09.
Numbers in the bold indicate that the MSFE of the BVAR with Median CPI is less than the MSFE from the
BVAR without Median CPI.
*denotes significance at 5% level
**denotes significance at 10% level
(based on modified Diebold-Mariano test)

Table A4: Forecast Comparison of Benchmark BVAR in Carriero et al (2011) --- Pre-Crisis
Relative Mean Squared Error --- relative to modified BVAR5
A4a: BVAR5 vs. BVAR5 with median CPI inclusion
h=1M
0.995

h=6M
0.978

h=9M
1.010

h=12M
1.016

h=15M
1.017

h=18M
1.012

h=21M
1.008

h=24M
1.022

Core CPI

0.912***

0.961

0.938

0.948

0.939

0.939

0.933

0.918

Headline CPI

0.978

0.982

1.006

0.947

0.981

0.989

0.988

0.976

UR

1.004

0.981

0.988

0.987

0.984

0.992

0.997

0.998

Fed Funds Rate

1.026

0.935

0.940

0.924

0.908

0.898

0.892

0.878

Payroll growth

A4b: BVAR5 vs. BVAR5 with median CPI replacing core CPI
h=1M
1.008

h=6M
1.023

h=9M
1.017

h=12M
0.999

h=15M
0.991

h=18M
1.001

h=21M
1.009

h=24M
1.015

Core CPI
Headline CPI

----0.975

----0.969

----0.988

----0.935*

----0.975

----0.967

----0.978

----0.961*

UR

1.005

0.992

1.048

1.034

1.011

1.002

1.004

1.011

Fed Funds Rate

1.068

0.946

0.967

0.957

0.927

0.918

0.912

0.900

Payroll growth

Notes for the table: The tables A4a-A4b lists the mean squared forecast error (MSFE) of the BVAR with
Median CPI added to it relative to the mean squared forecast error of the BVAR without it. The reported
RMSFEs are for the payroll growth (monthly at annual rate), core CPI inflation (monthly at annual rate),
headline CPI inflation (monthly at annual rate), the unemployment rate, and the federal funds rate for
h=1,2,…24 step head forecasts (i.e. 2 years out) for the evaluation period 1987M1 – 2007M09.
Numbers in the bold indicate that the MSFE of the BVAR with Median CPI is less than the MSFE from the
BVAR without Median CPI.
*denotes significance at 5% level
**denotes significance at 10% level
(based on modified Diebold-Mariano test)

Table A5: Exercise using the median CPI to forecast PCE-based inflation using Carriero et al (2011)
Monthly BVAR--- Pre-Crisis
Relative Mean Squared Error --- relative to BVAR6
A5a: BVAR6 vs. BVAR6 with median CPI inclusion
Payroll growth

h=1M
1.003

h=6M
1.017

h=9M
1.049

h=12M
1.015

h=15M
0.987

h=18M
0.992

h=21M
1.010

h=24M
1.032

Core PCE inflation

0.965

0.972

0.959

0.968

0.931*

0.907** 0.894** 0.872**

Headline PCE inflation

0.961

0.955

0.972

0.949

0.943

0.923**

0.934*

0.911**

UR

1.005

0.988

1.023

1.026

1.003

0.987

0.982

0.982

Fed Funds Rate

1.010

0.918**

0.970

0.960

0.927

0.904

0.885

0.862*

A5b: BVAR6 vs. BVAR6 with median CPI replacing core PCE
Payroll growth

h=1M
0.987

h=6M
1.003

h=9M
1.043

h=12M
1.005

h=15M
0.984

h=18M
1.006

h=21M
1.019

h=24M
1.030

Core PCE inflation
Headline PCE inflation

----0.982

----0.951

----0.967

----0.943

----0.939

----0.917**

----0.932*

----0.916**

UR

1.017

0.961

1.001

1.000

0.968

0.955**

0.964

0.979

Fed Funds Rate

1.007

0.902*

0.953

0.933

0.889

0.867

0.851*

0.832*

Notes for the table: The tables A5a-A5b lists the mean squared forecast error (MSFE) of the BVAR with
Median CPI added to it relative to the mean squared forecast error of the BVAR without it. The reported
RMSFEs are for the payroll growth (monthly at annual rate), core PCE inflation (monthly at annual rate),
headline PCE inflation (monthly at annual rate), the unemployment rate, and the federal funds rate for
h=1,2,…24 step head forecasts (i.e. 2 years out) for the evaluation period 1987M1 – 2007M09.
Numbers in the bold indicate that the MSFE of the BVAR with Median CPI is less than the MSFE from the
BVAR without Median CPI.
*denotes significance at 5% level
**denotes significance at 10% level
(based on modified Diebold-Mariano test)

Table A6: Exercise using the median CPI to forecast PCE-based inflation using Beauchemin and Zaman
(2011) Quarterly BVAR--- Pre-Crisis

Relative Mean Squared Error (RMSE) --- relative to BVAR7
A6a: BVAR7 vs. BVAR7 with median CPI inclusion
Real GDP growth

h=1Q
0.997

h=2Q
1.000

h=3Q
1.018

h=4Q
1.009

h=5Q
1.004

h=6Q
1.010

h=7Q
1.011

h=8Q
1.008

Core PCE inflation

0.973

0.960

0.962

0.953

0.946*

0.942*

0.945*

0.953

Headline PCE inflation

0.996

0.982

0.978

0.957

0.957

0.960*

0.957*

0.961

Unemployment Rate

1.001

0.999

1.004

1.009

1.008

1.008

1.013

1.019

Fed Funds Rate

1.007

0.994

0.996

0.995

0.987

0.980

0.972*

0.970*

A6b: BVAR7 vs. BVAR7 with median CPI replace core PCE
Real GDP growth

h=1Q
0.992

h=2Q
1.000

h=3Q
1.016

h=4Q
1.010

h=5Q
1.003

h=6Q
1.004

h=7Q
0.999

h=8Q
0.994

Core PCE inflation
Headline PCE inflation

----0.998

----0.987

----0.983

----0.961

----0.965

----0.974

----0.972

----0.981

Unemployment Rate

0.983

0.977

0.987

0.993

0.995

0.997

1.003

1.009

Fed Funds Rate

0.998

0.974

0.971

0.966

0.960*

0.959*

0.956*

0.958

Notes for the table: The tables A6a-A6b lists the mean squared forecast error (MSFE) of the BVAR with
Median CPI added to it relative to the mean squared forecast error of the BVAR without it. The reported
RMSFEs are for the real GDP growth (quarterly at annual rate), core PCE inflation (quarterly at annual
rate), headline PCE inflation (quarterly at annual rate), the unemployment rate, and the federal funds
rate for h=1,2,…8 step head forecasts (i.e. 2 years out) for the evaluation period 1987Q1 – 2007Q3.
Numbers in the bold indicate that the MSFE of the BVAR with Median CPI is less than the MSFE from the
BVAR without Median CPI.
*denotes significance at 5% level
**denotes significance at 10% level
(based on modified Diebold-Mariano test)

Table A7: Exercise using the forecasted values of the median CPI as the forecast for core CPI with
Beauchemin and Zaman (2011) Quarterly BVAR--- Pre-Crisis

Relative Mean Squared Error (RMSE) --- relative to BVAR1
A7a: BVAR1 vs. BVAR1 in which we use Median CPI to predict Core CPI

Real GDP growth

h=1Q
1.023

h=2Q
1.011

h=3Q
1.006

h=4Q
1.006

h=5Q
0.996

h=6Q
1.007

h=7Q
1.006

h=8Q
0.996

Core CPI inflation

1.114

1.095

0.996

0.937

0.915

0.896

0.903

0.913

Headline CPI inflation

0.966

0.960

0.964

0.938

0.938

0.954

0.959

0.966

Unemployment Rate

1.018

1.011

1.030

1.027

1.022

1.018

1.014

1.011

Fed Funds Rate

0.986

0.978

0.974

0.966

0.958

0.955

0.947*

0.946*

Notes for the table: The table A7a lists the mean squared forecast error (MSFE) of the BVAR (which is
estimated using Median CPI and uses the forecasts of the Median CPI to predict core CPI) relative to the
mean squared forecast error of the BVAR without Median CPI. The reported RMSFEs are for the real
GDP growth (quarterly at annual rate), core CPI inflation (quarterly at annual rate), headline CPI inflation
(quarterly at annual rate), the unemployment rate, and the federal funds rate for h=1,2,…8 step head
forecasts (i.e. 2 years out) for the evaluation period 1987Q1 – 2007Q3. Numbers in the bold indicate
that the MSFE of the BVAR with Median CPI is less than the MSFE from the BVAR without Median CPI.
*denotes significance at 5% level
**denotes significance at 10% level
(based on modified Diebold-Mariano test)

Table A8: Exercise using the forecasted values of the median CPI as the forecast for core CPI with
Carriero et al (2011) Monthly BVAR--- Pre-Crisis

Relative Mean Squared Error --- relative to BVAR5
A8a: BVAR5 vs. BVAR5 in which Median CPI is used to predict core CPI
Payroll growth

h=1M
0.973

h=6M
1.020

h=9M
1.015

h=12M
0.996

h=15M
0.988

h=18M
0.997

h=21M
1.006

h=24M
1.013

Core CPI

0.907

0.954

0.904

0.984

0.996

0.954

0.900

0.848

Headline CPI

0.983

0.969

0.994

0.937*

0.969

0.964

0.976

0.965

UR

1.001

0.989

1.046

1.033

1.012

1.003

1.005

1.012

Fed Funds Rate

1.052

0.945

0.966

0.955

0.927

0.920

0.915

0.903

Notes for the table: The table A8a lists the mean squared forecast error (MSFE) of the BVAR (which is
estimated using Median CPI and uses the forecasts of the Median CPI to predict core CPI) relative to the
mean squared forecast error of the BVAR without Median CPI. The reported RMSFEs are for the payroll
growth (monthly at annual rate), core CPI inflation (monthly at annual rate), headline CPI inflation
(monthly at annual rate), the unemployment rate, and the federal funds rate for h=1,2,…24 step head
forecasts (i.e. 2 years out) for the evaluation period 1987M1 – 2007M09. Numbers in the bold indicate
that the MSFE of the BVAR with Median CPI is less than the MSFE from the BVAR without Median CPI.

*denotes significance at 5% level
**denotes significance at 10% level
(based on modified Diebold-Mariano test)