ShortRun Restrictions: An Identification Device Fabrice Collard, Patrick Fve

1
Short-Run Restrictions An Identification
Device?Fabrice Collard, Patrick Fève Julien
MatheronDiscussion

• Gert Peersman
• Ghent University

2
Situation in literature

• Very interesting paper
• Estimation of deep parameters of DSGE models
• Strong econometric interpretation
• Full characterisation of the data
• Maximum likelihood methods or Bayesian methods
• Weak econometric interpretation
• DSGE model is calibrated as such that theoretical
  moments match their counterparts in the data
• Fully specifying the stochastic structure of the
  model is not required
• Often more robust than full-information
  estimators
• E.g. minimising the difference between an impulse
  response function from an empirical VAR and a
  theoretical DSGE model

3
Situation in literature

• Weak econometric interpretation
• Minimum distance estimation where structural
  shocks in the VAR are identified with short-run
  zero restrictions
• Rotemberg and Woodford (1998)
• Christiano, Eichenbaum and Evans (2005)
• Same restrictions are required in DSGE model
• This paper analyses whether this approach can be
  used to identify the deep parameters of the DSGE
  model

4
But...

• RW (1998) and CEE (2005) identify monetary policy
  shocks with short-run zero restrictions while
  this paper identifies a technology shock to
  evaluate the method
• Impact of techology shock on labour (a response
  to be matched) is already very controversial in
  literature
• Are the same conclusions also found when a
  monetary policy shock is identified with
  short-run restrictions?
• The way short-run restrictions are implemented to
  identify technology shocks in the VAR using real
  data is problematic
• Technology shock is a shock with no immediate
  effect on hours in a VAR containing hours and
  labour productivity
• Implies that all other shocks do have an
  immediate impact on hours and labour decisions
  (including e.g. monetary policy shocks)
• Identified shock in paper will be a combination
  of several shocks

5
But...

• The way short-run restrictions are implemented to
  identify technology shocks in the VAR using real
  data is problematic
• Technology shocks are usually identified with
  long-run restrictions
• Authors show that impulse responses look similar
  anyway
• Based on VAR estimated with hours in levels
  (Christiano, Eichenbaum and Vigfusson 2005a)
  while original papers are estimated with first
  difference specification (Gali 1999 or Francis
  and Ramey 2002)
• Estimated impulse response functions using
  long-run restrictions are very uncertain never
  significant different from zero (not only
  contemporaneous impact)
• Short-run restrictions to identify technology
  shocks also done by Christiano, Eichenbaum and
  Vigfusson (2005b)
• But they use simulated data obtained from a model
  containing the restriction with only one shock

6
Contribution of this paper

• CEE (2005) monetary policy shock
• Robustness check the role of timing assumptions
• Eyeball econometrics by having a look at
  responses when some variables are not
  predetermined
• This paper adds some nice supplements (IRF ratio,
  autocorrelation function) and also a statistical
  procedure
• Application to impact of technology shock
• Information structure does matter!
• Question the use of short-run zero restrictions
  to identify deep parameters of DSGE model!

7
Going even further

• Short-run zero restrictions to identify the
  shocks in the VAR is already very controversial
• CEE (2005) long standing view that many
  macroeconomic variables do not respond
  immediately to policy shocks
• Start introducing these restrictions in their
  theoretical model
• these assumptions do not have a substantial
  impact on dynamic properties of the model
  (magnitude, persistence, hump-shaped response)
• No theoretical reason to believe in short-run
  zero restrictions (see also Canova and Pina,
  2005)
• Short-run zero restrictions have a considerable
  effect on the estimated impact of shocks in VARs
  (Peersman, 2005)

8
Going even further

• Response to a monetary policy shock of 50 basis
  points

• Response function with zero restrictions in tails
  of all possible responses
• Also variance decompositions changes dramatically
• What happens with deep parameters if these more
  general restrictions are implemented in the VAR?

9
Going even further

• Also long-run zero restrictions can be
  controversial
• Often questioned from an empirical point of view
  Sims (1972), Faust and Leeper (1997), Erceg,
  Guerrieri and Gust (2005), ...
• Farrant and Peersman (2006) contribution of
  nominal shocks to the euro/dollar exchange rate
  increases from 11 to 57 when qualitative
  restrictions are used instead of long-run zero
  restrictions (both obtained from the same
  theoretical model)
• Peersman and Straub (2004) Sign of the impact of
  technology shocks on hours changes with
  qualitative restrictions

10
Going even further

• Go even more general if some of qualitative
  restrictions are uncertain or want to be tested
• Peersman and Straub (2006) use sign restrictions
  on the ratio of responses
• Example many models predict a rise in output and
  a fall in private consumption and investment
  after a positive government spending shock
• Not useful restrictions not robust for all
  models and often rejected in empirical papers
• Impose a much more general restriction on the
  consumption-output and investment-output ratio
• Data can then determine the reaction of
  consumption and investment

11
Going even further

• Peersman and Straub (2006) use sign restrictions
  on the ratio of responses
• Expansionary government spending shocks generate
  a fall in private consumption and investment
• A positive shock in consumer preferences has a
  positive effect on investment (in contrast to New
  Keynesian DSGE models)
• A favourable shock in investment adjustment costs
  has a positive impact on private consumption (in
  contrast to New Keynesian DSGE models)
• Technology shocks cause hours to rise (in
  contrast to New Keynesian DSGE models)