Discussion of “Learning to Set Policy Optimally” by Richard Dennis and Federico Ravenna Ulf S¨oderstr¨om IGIER, Universit`a Bocconi December 2004
Ulf S¨ oderstr¨om
Discussion of Dennis and Ravenna
What’s this paper all about? Central banks must set policy, but don’t know the true model.
Try to learn over time. Typically: Focus on private sector learning, CB uses Taylor-type rule
⇒ Policy independent of learning process. Here: CB learns parameters, sets policy optimally (targeting rule)
⇒ Learning affects policy rule. Main questions:
– Is REE learnable? (With and without private sector learning and model misspecification.) – What are the welfare consequences of learning?
1
Ulf S¨ oderstr¨om
What do they find? REE learnable if CB uses correct model. REE not learnable if CB’s model misspecified. But: Welfare costs small. Even: Welfare benefits from learning.
Discussion of Dennis and Ravenna
2
Ulf S¨ oderstr¨om
Discussion of Dennis and Ravenna
What do I think? Several nice points:
– Optimal policy affected by learning. – Non-optimizing central bank may learn faster. – Not only learnability, also welfare consequences. Main comments:
– Learnability with misspecified model – Welfare measure – Policy assumption
3
Ulf S¨ oderstr¨om
Discussion of Dennis and Ravenna
Learnability REE learnable if CB uses correct model
Not learnable if misspecified model. In practice: CB knows model is misspecified, but not how. Relevant question: Will CB ever realize that the model is misspecified? Here: CB never suspects the model is misspecified.
4
Ulf S¨ oderstr¨om
Discussion of Dennis and Ravenna
5
Welfare measure If policy optimal, learning affects policy rule. But: Ad hoc loss function to set policy and evaluate welfare. If welfare-based objectives, learning affects also objectives. Walsh (2004): welfare effects of parameter uncertainty smaller if true welfare
measure used to set policy. Cheap point, but may be important here.
Ulf S¨ oderstr¨om
Discussion of Dennis and Ravenna
6
Policy At first glance, striking result: Learning may improve on welfare.
– Why? Discretionary policy suboptimal. – Cf. policy with unobservable states (Nimark, 2003; Lippi and Neri, 2004) With discretionary policy, 2 effects of learning:
– Learning reduces welfare (because CB uses wrong parameter values) – Learning may improve on discretionary policy (because policy not optimal) Net effect sometimes positive, sometimes negative. Can these effects be disentangled?
Ulf S¨ oderstr¨om
Discussion of Dennis and Ravenna
7
Policy Why not commitment?
– Use welfare-maximizing policy as benchmark (full commitment, not timeless perspective). – Paper is on costs of learning, not costs of discretion (Dennis and S¨oderstr¨om, 2004). – Discretion may be more reasonable, but distorts welfare effects of learning. Also: Will non-optimizing CB learn faster? Compare with Taylor rule.
Ulf S¨ oderstr¨om
Discussion of Dennis and Ravenna
Where to go next? Characterize policy under learning. More inertial? Explains CB inertia? (Improves on discretion.) “MP inertial because CB learns faster that way”:
Not realistic description of policy process (cf. Blinder, 1998, on experimentation) Rudebusch (2002) puzzle: Persistence but not predictability.
– At any point in time, responses depend on parameter estimates ⇒ not much inertia? – Over time, parameter estimates evolve ⇒ additional inertia?
8
Ulf S¨ oderstr¨om
Discussion of Dennis and Ravenna
9
References Blinder, Alan S. (1998), Central Banking in Theory and Practice, The MIT Press. Dennis, Richard and Ulf S¨oderstr¨om (2004), “How important is precommitment for monetary policy?” Manuscript, IGIER, Universit`a Bocconi. Lippi, Francesco and Stefano Neri (2004), “Information variables for monetary policy in a small structural model of the euro area,” Working Paper No. 511, Banca d’Italia. Nimark, Kristoffer P. (2003), “Indicator accuracy and monetary policy: Is ignorance bliss?” Working Paper No. 157, Sveriges Riksbank. Rudebusch, Glenn D. (2002), “Term structure evidence on interest rate smoothing and monetary policy inertia,” Journal of Monetary Economics, 49 (6), 1161–1187. Walsh, Carl E. (2004), “Parameter misspecification and robust monetary policy rules,” Manuscript, University of California, Santa Cruz.