Forward Guidance and Long Term Interest Rates: Inspecting the Mechanism

Transcription

1 Forward Guidance and Long Term Interest Rates: Inspecting the Mechanism Ferre De Graeve Pelin Ilbas Raf Wouters Rennes, December 4-5, 14 De Graeve, Ilbas & Wouters Forward () guidance & long rates 1 / 43

2 Forward guidance Announcement of future policy intentions Time-dependent State-dependent In ation targeting CB s Transparency about policy rule Zero lower bound Stimulus beyond usual policy rule / 43

3 Communicated transmission: Lower long maturity interest rates Fed statements: vague, no distinction between nominal and real (long) rates Fed reports:... reiteration by the FOMC that it expected to maintain exceptionally low levels of the federal funds rate for an extended period... contributed to the decline in nominal Treasury yields (MPR, July 11) Empirical evidence: event studies measure fall in nominal long rate e.g. Willliams, 13; Woodford, 13; Moessner, 13 3 / 43

4 Source: Williams, / 43

5 Outline Detail transmission of: announced monetary policy shock time-dependent FG announced shock in the presence of a constant interest rate path time-dependent FG at ZLB Do this in models of varying complexity Also: Separate di erent channels Which model-features/parameters matter for di erent channels? Which characteristics of transmission are robust? State-dependent FG Incomplete information e Graeve, Ilbas & Wouters Forward ( guidance Sveriges& Riksbank long rates & NBB) 5 / 43

6 Preview of results. DSGE models provide useful insight into the transmission mechanism of forward guidance annnouncements. 1 The IRFunctions behave normal (no excessive response or sign reversals) if credible policy scenario s are considered, which means that policy announcement are formulated as state dependent intentions. 6 / 43

7 Preview of results For successful forward guidance, a fall in nominal long rate is 1 not necessary The reaction of the nominal long rate to forward guidance expansions depends on the endogenous persistence in output, in ation and the standard policy reaction: the long rate will react positive if the economy is still booming and in ation rising during the exit period. not su cient Imperfect information about the rationale behind the policy announcement can imply both a fall in the long rate and a much attenuated stimulus e ect on the real economy. 7 / 43

8 Model (FWD) 1 y t = y t+1 σ [r t π t+1 ] (1 θ)(1 θβ) π t = βπ t+1 + i θ r t = hφ π π t + φ y y t + ε r t + ε r1 t r 1y t = i= E t r t+i y t ε rn t n Forward Guidance: r FG, r1 FG,..., rn FG ( i hφ π π t + φ y y t r t = r FG t if t > n otherwise. 8 / 43

9 Model (FWD) Consider ε r8 t 8 < 9 / 43

10 1 SHORT RATE.5 LONG RATE REAL LONG RATE OUTPUT INFLATION SHOCK Announced shock 1 / 43

11 Constrained path + announced shock (FWD) As illustration of ZLB binds + prolongation of staying there 11 / 43

12 1 SHORT RATE.5 LONG RATE REAL LONG RATE OUTPUT INFLATION SHOCK Announced shock Announced shock + path 1 / 43

13 Model (FWD & BWD) h y t = (1 + h) y t (1 + h) y (1 h) t+1 σ(1 + h) [r t π t+1 ] + ε y t ι π t = (1 + ιβ) π β t 1 + (1 + ιβ) π (1 θ)(1 θβ) t+1 + y t θ(1 + ιβ) i r t = ρr t 1 + (1 ρ) hφ π π t + φ y y t + ε r t + ε r1 t ε rn t r 1y t = i= E t r t+i Sources of persistence: h, ι, ρ (φ y ) n 13 / 43

14 .5 SHORT RATE. LONG RATE REAL LONG RATE INFLATION OUTPUT SHOCK Announced shock Announced shock + path e Graeve, Ilbas & Wouters Forward ( guidance Sveriges& Riksbank long rates & NBB) 14 / 43

15 1 SHORT RATE.5 LONG RATE REAL LONG RATE OUTPUT INFLATION SHOCK Announced shock Announced shock + path e Graeve, Ilbas & Wouters Forward ( guidance Sveriges& Riksbank long rates & NBB) 15 / 43

16 Long rate response So far discussed the channels What determines their strength, and thus the long rate response? Stronger MTM (for given MP params) =) strength boom increases =) endogenous channel becomes stronger =) long rate more likely to increase Stronger policy response (for given MTM) =) endogenous channel becomes stronger =) long rate more likely to increase (but o setting dynamics in in ation dynamics) More persistence (ι, h, ρ, φ y ) =) post-implementation endogenous response more important 16 / 43

17 h=, ι= LONG RATE h=, ι=.5 h=, ι= h=.45, ι= 1 h=.45, ι=.5 1 h=.45, ι= h=.85, ι= 1 h=.85, ι=.5 1 h=.85, ι= / 43

18 OUTPUT 1 h=, ι= 1 h=, ι=.5 1 h=, ι= h=.45, ι= h=.45, ι= h=.45, ι= h=.85, ι= h=.85, ι= h=.85, ι= / 43

19 h=, ι= REAL LONG RATE h=, ι=.5 h=, ι= h=.45, ι= h=.45, ι= h=.45, ι= h=.85, ι= h=.85, ι= h=.85, ι= / 43.1

20 Channels 1 Depending on various params/frictions, nominal long rate response can be all over the place Despite FG working 3 What matters is long real rate (a more robust feature) / 43

21 FG in medium-scale DSGE models Of interest, since Academic: FG-puzzle (DGP): macro-impact too high, long rate impact implausibly low Used for policy 1 / 43

22 1 short rate.1 long yield output 3 inflation real long yield GSW DGP SW / 43

23 e Graeve, Ilbas & Wouters Forward ( guidance Sveriges& Riksbank long rates & NBB) 3 / 43

24 State-dependent FG reasons: Actual policy FOMC switched from date-based to state dependent forward guidance in December 1 by announcing that no increase in the federal funds rate should be expected at least as long as the unemployment rate remains above 6 1/ percent, in ation between one and two years ahead is projected to be no more than a half percentage point above the Committee s percent longer-run goal Potential resolution to excess sensitivity à la Carlstrom et al. (1) Less extreme policy experiments 4 / 43

25 State-dependent FG Example: GSW (which features unemployment) 4 scenarios, all start from the observed state in 1Q4, i.e. the start of state-dependent FG 1 no zlb zlb 3 zlb + state FG 4 zlb + state FG + 1Q additional zlb 5 / 43

26 6 short rate 6 long yield output 4 inflation real long yield 1 unemployment / 43

27 6 short rate 6 long yield output 4 inflation real long yield 1 unemployment / 43

28 6 short rate 6 long yield output 4 inflation real long yield 1 unemployment / 43

29 8 short rate 6 long yield output 1 inflation real long yield 1 unemployment / 43

30 Imperfect information Often discussed (e.g. Campbell et al. 1, Filardo and Ho man, 14) Uncertain interpretation of announced FG: Is it... 1 additional exogenous stimulus for given state of the economy? endogenous response to worse state of the economy? 3 / 43

31 Imperfect information Setting: (symmetric) partial information - Pearlman, Currie & Levine (1986) Agents do not observe shocks outright, but do observe fr, π, w, y, i, c, lg They understand e.g. the short rate will fall, but not necessarily why Note 1: conduct all experiments in SW Note : in SW, these observables su ce to retrieve full info IRF 31 / 43

32 General implications of imperfect information 1 Confusion =) updating EH still holds (but expectations may be o ) 3 Long rate accumulates confusion (at each point in time) Remainder: SW-speci c implications of imperfect information 3 / 43

33 Unanticipated shocks still very well identi ed. Short rate. Long rate Inflation Real long rate Full info Partial info Output De Graeve, Ilbas & Wouters Forward ( guidance Sveriges& Riksbank long rates & NBB) 33 / 43

34 Filtering problem most pertinent with anticipated shocks.1 Short rate. Long rate Inflation Real long rate Full info Partial info Output De Graeve, Ilbas & Wouters Forward ( guidance Sveriges& Riksbank long rates & NBB) 34 / 43

35 Confusion disappears after a while: when the shock arrives (not always) Short rate: Full info Short rate: Partial info.4.4. Actual.. E 1 x t..4.4 E.6 x t Inflation: Full info Output: Full info.3. Inflation: Partial info Output: Partial info De Graeve, Ilbas & Wouters Forward ( guidance Sveriges& Riksbank long rates & NBB) / 43

36 examples that hint at ZLB scenarios Have not implemented xed rate path under imperfect information Less active policy, 1Q ahead announcement φ π = 1.1 Predetermined policy rule: lagging policy reaction =~ ZLB irf under FI i r t = ρr t 1 + (1 ρ) hφ π (E π t+6 ) t 6 + φ y (Ey t+6 ) t 6 + ε r t / 43

37 Imperfect information - less active policy rule Short rate: Full info Short rate: Partial info.5 Actual E 1 x t.5 E.1 x t Inflation: Full info Output: Full info Inflation: Partial info Output: Partial info / 43

38 .1 Short rate.5 Long rate Inflation Real long rate.1 Full info Partial info Output / 43

39 Imperfect information - xed short rate (predetermined policy rule).5 Short rate: Full info Actual.5 E 1 x t.1 E x t Inflation: Full info Output: Full info Short rate: Partial info Inflation: Partial info Output: Partial info / 43

40 . Short rate. Long rate Inflation Real long rate Output. Full info Partial info / 43

41 examples that hint at ZLB scenarios =) Both generate reduction in long rate, lesser boom Implication: Change in the long rate is 1 particularly sensitive (it accumulates uncertainty over many horizons) especially on impact (when most of the updating still has to take place) 3 yet that is exactly what is being measured in event studies. 41 / 43

42 In sum Fall in long rate 1 Not necessary: many models imply increase due to endogenous leaning against boom Not su cient: imperfect infomation examples suggest plausibility of confusing exogenous FG with bad news assessment (will imply lesser boom, lower long rate) 3 Not what is suggested by all empirical DSGE models * 3-eq NK which replicates unanticipated MP IRF * (G)SW (contrary to DGP, not ne-tuned) 4 / 43

43 Implications Event-study evidence (used in the evaluation of FG): 1 inference exactly when confusion is largest di cult to interpret (not clear what can be learned from it, since long rate not nec./suf.) Communication: 1 more info: transparency on policy intentions can avoid confusion more focus on the real rate (closer to the actual objective + more robust across models) 43 / 43