Transcript Slides:PPT - Tony Yates

Structural VARs: celebrated
applications and commentaries
Revision Lecture for MSc Time Series
Econometrics module, Bristol, Spring
2014
Overview
• Sims on the ‘price puzzle’
• Christiano et al on identifying mon pol shock using
Cholesky
• Gali on identifying technology shocks using long run
restrictions
• Francis et al/Barsky-Sims on max share restrictions and
news shocks
• Canova and de Nicolo on identifying the output effects of
monetary policy shocks using sign restrictions
• Rigobon on identification through heteroskedasticity
• Pinter et al [inc me] on sign restrictions, Bayesian VARs,
news shocks
Overview (2)
• Rudebusch vs Sims on monetary policy
‘shocks’.
• Romer and Romer: narrative methods for
identifying policy shocks.
• Stock and Watson on a factor analysis of the
financial crisis. [See earlier lecture slides for
detailed look].
What to get from a paper for the exam
[and life!]
• What did the authors do and why? What
debate were they contributing to?
• What did they find? How did this change
what we thought about the debate they were
contributing to?
• What further work did it prompt? What
criticisms can be made? What questions did
their work beg?
Sims and the price puzzle
• Cholesky identification often found that rise in rates
caused rise in prices. A puzzle from perspective of
modern macro models.
• Sims puts this down to missing variables, like
commodity prices. Fed responding to many variables,
not just inflation and GDP.
• Rise in rates looks exogenous, uncorrelated with
today’s inflation and GDP, but actually responding to
commodity prices, which may be a good forecast of
future GDP.
• Other explanations by later researchers offered too.
Sims
• Implication was you needed to enlarge VAR [in
later work with Zha, he tried 19 variables!].
• This spurs work on techniques to cope with
curse of dimensionality: Bayesian VARs, and
factor modelling.
CEE on cholesky identification of mon
pol shocks
• Find monetary policy has persistent and hump-shaped
responses on real variables, and inflation.
• Needs sticky prices/wages, indexation, habits, and
other rigidities to match.
• Inconsistent with RBC model [mp neutral on output],
and with simplest NK models [IRFs not persistent or
hump-shaped].
• Issue: indexation not consistent with micro data on
prices.
• View paper in light of Rotemburg-Woodford’s effort to
fit a simpler sticky price model to similarly identified
monetary policy shock responses.
Gali on technology shocks and long
run restrictions
• RBC crowd claimed that technology shocks
were dominant explanation for business
cycles.
• Gali used RBC/NK theory to identify
technology shocks; should be only thing
affecting labour productivity in long run.
• Saw that these shocks caused fall in hours
work. [=‘relax/don’t make hay while the sun
shines’]
Gali: ctd…
• Implied tech shocks not dominant cause of
business cycles: unconditional correlation of
GDP and hours was positive.
• Or implied prices were not flexible. Note
sticky prices means hours fall after positive
technology shock.
• Prompted furious debate about the efficacy of
long run restrictions in US academia.
Gali (3)
• Christiano et al showed that results depended on
the hours worked variable used. Shd it be hours,
or hours/head? How should it be de-trended?
• Giraitis et al show that there was time variation in
response of hours worked to tech shock. Became
more RBC like as time went by.
• Debate produced Francis et al’s paper on maxshare restrictions.
• Controversy over the sign of response of hours
worked implies difficult to use sign restrictions
[on hours] to identify this shock!
Francis et al on max share restrictions
• Critics of LR restrictions argued that finite sample
not appropriate for identifying the long run.
• Recall that LR restrictions built from infinite series
sum, involving ever higher powers of VAR
coefficients. Estimated with error.
• Francis et al suggest identifying shock by
‘rotating’ the Chol factor of the RF VCOV matrix
to max the share of variance at some long but
finite horizon.
Francis et al (2)
• They redo Gali’s analysis of technology shocks.
• Find that a technology shock increases hours
worked, in line with RBC, and contrary to
Gali’s original findings.
• Idea derives from papers by Faust (1996) and
Faust and Leeper (1997), explaining the pitfalls
of long run restrictions.
Canova and de Nicolo on sign
restrictions
• Identifies monetary policy shock based on sign
restrictions.
• Finds these shocks contribute significantly to
fluctuations in inflation and output.
• Latter indicates sticky prices.
• Some evidence of time variation.
• Other foundational papers on sign restrictions
are Faust (1998) and Uhlig (1999).
CN vs Uhlig
• CN: use sign of mon pol shock on ir, m/p [real
balances] and output. Finds MP contributes
great deal to output volatility. [<=60%]
• Uhlig: restricts responses of ir, m, p. Finds
mon pol shock has no significant effect on
output [implies prices are flexible].
Barsky-Sims on news shocks
• Exploits Francis et al’s idea on max share restrictions.
• Find innovation that is i) orthogonal to a proxy for tfp
today but ii) contributes maximally to tfp tomorrow.
• Depends on having found a proxy for tfp in the first
place, which is somewhat dubious.
• Find that news shocks are an important driver of the
business cycle.
• General implication being that business cycle diagnosis
can’t focus only on contemporaneously-revealed
shocks.
Pinter, Theodoridis and Yates on risk
news shocks
• Study of ‘risk news’ shocks. These are
revelations about future changes in the
variance of returns across different activities.
• Exploits Barsky-Sims’ method. News shock
constructed to be orthogonal to today’s risk
proxy, but contribute maximally to
fluctuations in the future.
• Identifies monetary policy, technology shocks
using sign restrictions at the same time.
Pinter et al (2)
• Paper is a VAR-based reflection on Christiano,
Motto, Rostagno, which claimed that risk
(contemp +news) shocks accounted for 60% of
business cycle volatility in output.
• Pinter et al find a number like 20%.
• Just as with Barsky-Sims, rather depends on
having found a good proxy for cross-sectional
risk.
Pinter et al (3)
• PTY use Bayesian methods to estimate the reduced
form VAR.
• That’s because they have 10 variables, and 3 lags.=300
coefficients.
• Modification of ‘Minnesota’ priors due to Doan et al ,
by Banburra et al.
• Idea: ‘twist’ posterior towards model that each series
is a random walk.
• In original applications, shown to improve forecasting
performance of the VAR.
• In this application, eradicates volatile IRFs.
‘Narrative measures’ of monetary and
fiscal policy shocks
• Aims to circumvent problems with VARs, or
corroborate their findings.
• Romer and Romer: trawl historical records for
fiscal policy changes interpreted as exogenous
to the business cycle.
• Such shocks contractionary, and more so than
those estimated using VARs.
• RR repeat same exercise for mon pol shocks.
• Repeated by Cloyne for UK
Sims vs Rudebush on monetary policy
shocks
• Rudebusch: ‘shocks’ don’t correlate well with
‘surprises’ measured by Fed Funds Futures.
• Sims: these surprises mix up forecast errors
due to genuine policy shocks, and errors due
to other shocks, which prompted a policy
response.
Stock and Watson very briefly
• Study financial crisis using very large panel of
200+ data series.
• Estimate that there are about 8 factors.
– Not very few as first thought.
– Early factor models had made RBC like claims that
only needed a few ‘drivers’ to explain business cycle.
• No new factors needed to explain post crisis
period.
– Crisis was larger version of old shocks, not a new
shock.