Discussion of Nakamura and Steinsson’s
“High Frequency Identification of Monetary
Non-Neutrality”
Eric T. Swanson
University of California, Irvine
ASSA Meetings, Boston
January 3, 2015
What the Paper Does
1. Estimates impulse response functions of interest rates and
inflation to an exogenous monetary policy shock.
2. Uses these impulse responses to estimate parameters of
the CEE model by matching model impulse responses to
empirical impulse responses.
Main contribution: Effects of monetary policy shocks are
identified using modern methods:
• high-frequency data
• heteroskedasticity (as in Rigobon 2003 REStat)
Identifying Effects of Monetary Policy Shocks
Canonical problem in empirical macro/monetary economics:
separate exogenous changes in monetary policy from
endogenous responses of monetary policy to the economy.
Modern approach: use daily or intra-daily data on days of
FOMC announcements
(Kuttner 2001 JME; Gurkaynak, Sack, Swanson 2005 AER, 2005 IJCB)
Identifying Effects of Monetary Policy Shocks
source: Gurkaynak, Sack, Swanson (2005 IJCB)
Identifying Effects of Monetary Policy Shocks
Canonical problem in empirical macro/monetary economics:
separate exogenous changes in monetary policy from
endogenous responses of monetary policy to the economy.
Modern approach: use daily or intra-daily data on days of
FOMC announcements
(Kuttner 2001 JME; Gurkaynak, Sack, Swanson 2005 AER, 2005 IJCB)
Caveat: some intermeeting FOMC announcements occurred
in response to weak employment reports, so daily data can be
endogenous.
Identifying Effects of Monetary Policy Shocks
source: Gurkaynak, Sack, Swanson (2005 IJCB)
Identifying Effects of Monetary Policy Shocks
Canonical problem in empirical macro/monetary economics:
separate exogenous changes in monetary policy from
endogenous responses of monetary policy to the economy.
Modern approach: use daily or intra-daily data on days of
FOMC announcements
(Kuttner 2001 JME; Gurkaynak, Sack, Swanson 2005 AER, 2005 IJCB)
Caveat: some intermeeting FOMC announcements occurred
in response to weak employment reports, so daily data can be
endogenous.
Potential endogeneity problem is even greater for two-day
event windows
(Gagnon, Raskin, Remache, Sack 2011 IJCB, Hanson and Stein 2012 FRBWP)
Identifying Effects of Monetary Policy Shocks
Two solutions to identification problem:
• intraday (tick) data
• identificaton through heteroskedasticity
(Rigobon 2003 REStat; Rigobon and Sack 2003 QJE, 2004 JME)
Nakamura and Steinsson use both of these methods.
However, there’s no incremental benefit to the
heteroskedasticity identification, once you have tick data.
In fact, heteroskedasticity-based formulas could contaminate
the estimates if non-FOMC homoskedasticity assumption is
violated.
Identifying Effects of Monetary Policy Shocks
Unconditional Volatility of 6-month T-bill Yield, 1990-2013
Identifying Effects of Monetary Policy Shocks
Two solutions to identification problem:
• intraday (tick) data
• identificaton through heteroskedasticity
(Rigobon 2003 REStat; Rigobon and Sack 2003 QJE, 2004 JME)
Nakamura and Steinsson use both of these methods.
However, there’s no incremental benefit to the
heteroskedasticity identification, once you have tick data.
In fact, heteroskedasticity-based formulas could contaminate
the estimates if non-FOMC homoskedasticity assumption is
violated.
Comment #1: Marginal benefit of heteroskedasticity
identification seems very low, given tick data.
One-Dimensional Monetary Policy?
Traditionally (Kuttner 2001 JME; Gurkaynak, Sack, Swanson 2005 AER)
change in monetary policy = change in fed funds rate
More recent literature (Gurkaynak, Sack, Swanson 2005 IJCB)
recognizes that monetary policy has two dimensions:
• changes in federal funds rate
• changes in forward guidance
Nakamura and Steinsson take one-dimensional approach, but
change in monetary policy ≠ change in fed funds rate
change in monetary policy ≠ change in forward guidance
Instead, change in monetary policy is some unspecified
average combination of the two.
One-Dimensional Monetary Policy?
source: Gurkaynak, Sack, and Swanson (2005 IJCB)
One-Dimensional Monetary Policy?
Empirically, changes in fed funds rate and changes in forward
guidance have different effects.
In DSGE models, changes in fed funds rate and changes in
forward guidance also have different effects
• contemporaneous shock vs. news shock
In Nakamura and Steinsson’s DSGE model, monetary policy
shocks are modeled in the traditional way—as a
contemporaneous shock to the federal funds rate.
Comment #2: There’s a mismatch between the empirical
impulse responses and the model’s impulse responses
• they are conceptually not the same type of shock
• model parameter estimates may be biased.
Risk Premia
Risk Premia
There’s some evidence that risk premia for very short-term
yields are small and relatively stable at daily frequency
(Piazzesi and Swanson 2008 JME)
But for longer-term bonds, risk premia are larger on average
and more volatile over time.
Nakamura and Steinsson consider some robustness checks:
• Blue Chip forecasts
• affine term structure model
• longer event-study windows
They conclude that the Expectations Hypothesis holds around
FOMC announcements.
Comment #3: Does the EH hold around FOMC (and/or other
major macroeconomic) announcements?
Summary of Comments
1. Given tick data, no need for heteroskedasticity-based
identification
• in fact, heteroskedasticity adjustment may contaminate
estimates
2. Assumption of one-dimensional monetary policy is
problematic
• empirical estimates are for an unspecified average of
changes in fed funds rate and forward guidance
• but monetary policy in DSGE model is modeled as a pure
change in fed funds rate—not the same thing!
3. Does the Expectations Hypothesis hold around FOMC (and
other major macroeconomic) announcements?