Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Discussion of D’Amico and King’s “What Does Anticipated Monetary Policy Do?” Eric T. Swanson University of California, Irvine ASSA Meetings San Francisco January 4, 2016 Summary of Paper Idea: get a model-free estimate of effects of forward guidance shocks on GDP, inflation, etc. Estimate a VAR on macro variables and survey expectations: Identify forward guidance shocks using sign restrictions: • lower expected future interest rates • higher expected future GDP • higher expected future inflation Compare results to forward guidance shocks in standard New Keynesian model Summary: Effects of Fwd Guidance are Big Authors find that forward guidance has large effects Conventional monetary policy: GDP Forward guidance: GDP Comment #1: Zero Restriction on Current yt, πt VAR Forward guidance should have same zero restrictions on output, inflation as conventional policy Comment #2: Sign Restriction on Current it VAR Residual covariance matrix Identify forward guidance shocks via sign restrictions: Comment #2: Sign Restriction on Current it Authors impose: A more natural restriction would be = 0 • (Same normalization as Gurkaynak, Sack, Swanson, 2005) • We’re interested in changes in forward guidance that are orthogonal to (i.e., above and beyond) changes in the current funds rate • Authors’ subsequent analysis always includes offsetting shocks to the current funds rate, anyway • During ZLB period, equality restriction clearly holds • Failure to impose this restriction may contaminate identification Comment #2: Sign Restriction on Current it Restriction matters: Interest rate expectations: Current interest rate: Comment #3: Including Forecasts in a VAR Including survey forecasts in a VAR creates inherent tension: • VAR is “true” law of motion for the economy • But survey forecast data provide alternate forecasts Authors assume Blue Chip survey forecasts are irrational To support this assumption, authors show VAR forecasts significantly outperform Blue Chip: But this result is • In sample • With Blue Chip forecasts included in the VAR • Including early 1980s (nominal data have downtrend) Comment #3: Including Forecasts in a VAR Compare to Faust and Wright (2013): “We find that judgmental survey forecasts outperform modelbased ones, often by a wide margin.” Overfitting seems to be a problem for the VAR Calls into question the estimated VAR reduced-form residuals, identification, and structural shocks: • Reduced-form residuals are too small • Structural shocks are too small • Covariance matrix Σ and hence Γ may not be representative • Structural shocks may not be well identified Comment #3: Including Forecasts in a VAR Seems to be a fundamental problem with VAR approach to incorporating “expectations shocks” Structural model-based analysis (as in NK-type models) seems better here: • Expectations shocks are well-defined in a structural model Authors’ goal of model-free estimates of effects of forward guidance may be too ambitious. Comment #4: ZLB Period is Problematic Zero lower bound period from 2009-15 is problematic: • ZLB violates linearity of the VAR • Unconventional monetary policy includes LSAPs as well as forward guidance LSAPs seem to satisfy the same sign restrictions as forward guidance Authors’ estimates of forward guidance shocks probably include LSAP effects as well Summary of Comments 1. Impose same contemporaneous zero restrictions on output and inflation for forward guidance as for funds rate 2. Impose restriction that forward guidance has no effect on current federal funds rate (instead of opposite effect) 3. Modeling “expectations shocks” in a VAR is problematic: • Drop GDP, inflation survey forecasts from VAR? • Structural NK-type models are probably a better way to model expectations shocks • Goal of model-free estimate of forward guidance effects may be too ambitious 4. ZLB period is problematic • focus on pre-2008 sample as baseline