Are Intrinsic Inflation Persistence Models Structural in the Sense of Lucas (1976)? Luca Benati, European Central Bank National Bank of Belgium November 19, 2008
This talk is based on 2 papers: Investigating Inflation Persistence Across Monetary Regimes (Quarterly Journal of Economics, 2008) Are Intrinsic Inflation Persistence Models Structural in the Sense of Lucas (1976)?, ECB, mimeo Motivation: Following Fuhrer and Moore (1995), several authors have proposed alternative mechanisms to hardwire inflation persistence into macro models, thus making it structural in the sense of Lucas (1976). Question: Are we sure that this is the right thing to do? My answer: Most likely not Hardwiring inflation persistence into macroeconomic models is wrong, and for policymaking purposes risky
The inflation persistence problem Point of departure, Fuhrer and Moore (1995): post-wwii U.S. inflation is very persistent traditional sticky-price models can t replicate such persistence Fuhrer and Moore s solution: the Fuhrer-Moore model, the first example of an intrinsic inflation persistence model: workers care about relative real wages (compared with previous and following cohorts of workers) this causes the Phillips curve to exhibit both a forward and a backward-looking component model-generated inflation is persistent Mechanism first introduced by Buiter and Jewitt (1981)
Following Fuhrer-Moore: intrinsic persistence models Following Fuhrer and Moore, several authors propose alternative mechanisms to hardwire persistence into macro models: hybrid i.e., mixed backward and forward-looking New Keynesian Phillips curves with indexation a-la-christiano- Eichenbaum-Evans, JPE 2005; Smets-Wouters, JEEA, 2003; etc ) information free-riders a-la-gali-gertler (JME, 1999) sticky information models Mankiw-Reis limited information processing capacity models Sims upward sloping hazard functions for the probability of a price change Sheedy (2007)
Most people work with U.S. post- WWII data, where inflation is indeed highly persistent But is post-wwii U.S. inflation representative of general characteristics inflation has exhibited through history? No! Two simple examples: My own perspective Neither under the Gold Standard nor under inflation targeting inflation was/is persistent. So, why should we hardwire inflation persistence in macro models??
Main results in these 2 papers: Under inflation targeting, in the U.K., Canada, Sweden, and New Zealand inflation exhibits little or no statistical persistence in the U.K. it is slightly negatively serially correlated the indexation parameter in hybrid New Keynesian Phillips curves is zero, or close to zero same result holds for alternative models with intrinsic inflation persistence (Gali-Gertler, Blanchard-Gali, ) Analogous evidence for Euro area, Germany, France, Italy, under European Monetary Union Under stable monetary regimes, both reduced-form and structural inflation persistence essentially disappear
Main results in these 2 papers (continued): Conceptually in line with Cogley-Sbordone (2006) Cogley and Sbordone (2006): if you control for trend inflation, inflation is purely forward-looking My work: - I compare inflation targeting, EMU, and Gold Standard where there is no trend inflation with regimes/ periods characterised by fluctuations in trend inflation - I find that inflation is (almost) purely forward-looking in the former regimes, but exhibits both statistical persistence and intrinsic persistence in the latter regimes - This questions the notion that intrinsic persistence is structural in the sense of Lucas (1976)
Reduced-form evidence for the United Kingdom Metallic standards: inflation was white noise or negatively serially correlated Interwar period: little persistence Bretton Woods: little persistence Bretton Woods to IT: very high persistence for much of this period, U.K. had no clear nominal anchor Inflation targeting: negative serial correlation
Current U.K. regime contains a component of mean reversion in the log price level it is a hybrid between inflation and price level targeting What about other inflation targeting countries? Evidence less dramatic than for the U.K., but: Bretton Woods to IT: generally high persistence Inflation targeting: very little persistence Finally but this is not new, see e.g. Barsky (1987) under the Gold Standard inflation was white noise in all countries I consider
What about Eurozone? Before EMU: a unit root process, possibly explosive After EMU: persistence has decreased in particular, for consumption deflator closer to targeted index, HICP, than GDP deflator inflation is estimated to be essentialy white noise
Critique: All this is purely reduced-form doesn t have any clear-cut implication for structural macro models What matters is structural inflation persistence i.e. a significant backward-looking component That s entirely correct, this evidence is suggestive that inflation might be purely forward-looking, but in no way it is decisive So let s go structural
Structural evidence E.g.: Linde (JME, 2005): if you use FIML, you get a dominant backward-looking component Linde s dataset: U.S., 1960Q1-1997Q4 Not surprising that he finds a dominant backward-looking component What if I apply full-information methods to EMU, inflationtargeting countries and data from the Gold Standard? I estimate via Bayesian methods a model very close to Linde Methodology: same as that of Schorfheide and co-authors random-walk Metropolis, etc, etc, etc Priors: all standard in the literature prior for indexation parameter flat over [0, 1] I want the data to speak freely
NK model with backward-and forward-looking components: IS shocks and monetary policy shocks allowed to be serially correlated Phillips curve shocks modelled as white noise I force all inflation persistence to be captured by indexation parameter, α I am stacking the cards against myself Let s see the results
Posterior distributions for the indexation parameter Full sample: high indexation for all countries Post-1999 Eurozone, IT countries, and Switzerland: backwardlooking component low to non-existent Post-1982 U.S.: consistent with Gali-Gertler (1999), inflation is less backward-looking Japan: even after the Great Inflation, still some intrinsic persistence
Gold Standard: U.S., U.K., Sweden, analogous results So, bottom line: Under stable monetary regimes, both reduced-form and structural inflation persistence vanish inflation is purely forward-looking These results question the notion that persistence found in U.S. post-wwii data is structural in the sense of Lucas (1976) They are compatible with the position of Cogley and Sbordone: controlling for trend inflation is key Critique: All this is based on models featuring indexation. What about other models featuring intrinsic persistence? Do you still get same results? Let s see
I consider four other models: Fuhrer and Moore (QJE, 1995) Gali and Gertler (JME, 1999) Blanchard and Gali (JMCB, 2007) Sheedy (2007, unpublished) All these models have one thing in common: within each model, there is one parameter encoding intrinsic persistence Question: Is this parameter invariant across regimes? No: under stable regimes with clearly-defined nominal anchors it often goes to zero
Example I: Blanchard and Gali (2007)
Example II: Sheedy (2007)
Implications Hardwiring post-wwii U.S. inflation persistence into the structure of the model is potentially highly misleading you estimate a structure which is not structural in the sense of Lucas (1976) What problems can you run into? You can t assess alternative monetary regimes in order to do that, you ought to have a model that is structural in the sense of Lucas (1976) suppose the U.S. Congress assessed the desirability of adopting an inflation (price level) targeting regime based on the model of Christiano-Eichenbaum-Evans chances are it would get the wrong answer
You can t compute optimal monetary policies Again, need a model that is structural in the sense of Lucas (1976) Suppose Bernanke computes an optimal monetary policy based on Christiano-Eichenbaum-Evans again, if I am right, he will get the wrong answer: optimal policy will be less aggressive than it should be