Econometric Analysis of Monetary Policy at the Zero Lower Bound - PowerPoint PPT Presentation

Econometric Analysis of Monetary Policy at the Zero Lower Bound Daisuke Ikeda Bank of Japan 26 December 2019 Shangshang Li (Oxford), Sophocles Mavroeidis (Oxford), and Francesco Zanetti (Oxford) The views expressed in this presentation are those of the author and should not be interpreted as those of the Bank of Japan. Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 1 / 33

Interest rates reached at the zero lower bound (ZLB) Short-term interest rate was a primary tool for monetary policy. Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 2 / 33

Unconventional monetary policy (UMP) Forward guidance (FG) – commitment about interest rates in future Quantitative easing (QE) – purchases of long-term government bonds Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 3 / 33

Two issues of monetary policy at the ZLB 1 Does the ZLB hamper the effectiveness of monetary policy? ZLB irrelevance hypothesis (e.g. Swanson and Williams 2014) 2 How effective is UMP under the ZLB? Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 4 / 33

Our approach: theory and evidence Theoretical model Simple New Keynesian model with: QE – long-term government bond purchases FG – keeping interest rates low for long Explains ZLB irrelevance hypothesis Empirical model Structural VAR (Mavroeidis 2019) with ZLB QE and FG in a similar spirit to the theoretical model Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 5 / 33

Main results 1 ZLB is empirically relevant for both Japan and the US ZLB irrelevance hypothesis is rejected 2 In the US, UMP has been quite (but not fully) effective Roughly 75% as effective as conventional one on impact Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 6 / 33

Related literature QE theory: Andres et al (2004); Chen et al. (2012); Harrison (2012); Gertler and Karadi (2013); Liu et al. (2019) FG theory: Reifschneider and Williams (2000) Empirical method: Mavroeidis (2019); Hayashi and Koeda (2019) ZLB irrelevant hypothesis: Swanson and Williams (2014); Debortoli et al. (2019) Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 7 / 33

Outline Macroeconomic model of UMP 1 Empirical model and identification at the ZLB 2 Testing ZLB irrelevance hypothesis 3 Impact of monetary policy 4 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 7 / 33

Model overview Based on 3-equation New Keynesian model Interest rate i t bounded below by 0 (ZLB) Shadow rate i ∗ t – the central bank’s “target” interest rate Depends on the Taylor-rule based rate and FG FG as in Reifschneider and Williams (2000) QE as in Chen et al. (2012) Bond market segmentation makes QE effective QE depends on i ∗ t Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 8 / 33

Model illustration Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 9 / 33

The model New Keynesian Phillips Curve y t − χ a z a ˆ π t = β E t ˆ π t +1 + κ ˆ t Euler equation, modified to incorporate QE � � y t +1 − 1 (1 − λ ∗ )ˆ i t + λ ∗ ˆ i ∗ − χ z z b y t = E t ˆ ˆ t − E t ˆ π t +1 t σ Interest rate rule, modified to incorporate FG � � � � t , − i ˆ ˆ i ∗ ˆ i ∗ t = ˆ i Taylor ˆ i t − ˆ i Taylor i t = max , − α , t t 1 + i ˆ i Taylor = ρ i ˆ i ∗ y t ) + ǫ i t − 1 + (1 − ρ i ) ( r π ˆ π t + r y ˆ t , t Notation: output ( y ); inflation ( π ); interest rate ( i ); shadow rate ( i ∗ ); Taylor-rule-based rate ( i Taylor ); productivity shock ( z a ); demand shock ( z b ); monetary policy shock ( ǫ i ). Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 10 / 33

Why do long-term rate and long-term gvt bonds disappear? QE under ZLB for long-term gvt bonds: ˆ b L , t = η i ∗ × i ∗ t . Long-term rate spread: ˆ ζ t = η b L × ˆ b L , t . Expected long-term rate: E t ˆ R L , t +1 = ... + η ζ × ˆ ζ t = > Both E t ˆ t +1 and ˆ R L b L , t can be replaced by i ∗ t . The efficacy of QE: λ ∗ ∝ η i ∗ × η b L × η ζ t ) when λ ∗ = 1 and α = 0 VAR(1) representation in ( π t , y t , i ∗ ⇒ ZLB is empirically irrelevant. Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 11 / 33

Effects of QE and FG Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 12 / 33

Effects of a monetary policy shock at the ZLB Figure: Impulse responses to a 1% increase in the interest rate at the ZLB ZLB caused by a severe demand shock Average of simulated responses 1000 times Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 13 / 33

Outline Macroeconomic model of UMP 1 Empirical model and identification at the ZLB 2 Testing ZLB irrelevance hypothesis 3 Impact of monetary policy 4 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 13 / 33

Empirical model (Mavroeidis 2019) Y 1 t = { inflation , output , long-term rate , ... } ; Y 2 t = policy rate. Y ∗ 2 t = shadow rate, representing desired policy stance C ensored and K inked S tructural VAR ( CKSVAR ): Y 1 t = β ( λ Y ∗ 2 t + (1 − λ ) Y 2 t ) + B 1 X t + B ∗ 12 X ∗ 2 t + ǫ 1 t , Y ∗ 2 t = − α Y 2 t + (1 + α ) ( γ Y 1 t + B 2 X t + B ∗ 22 X ∗ 2 t + ǫ 2 t ) , Y 2 t = max { Y ∗ 2 t , b t } where X t = { Y t − 1 , ..., Y t − p } and X ∗ 2 t = { Y ∗ 2 t − 1 , ..., Y ∗ 2 t − p } . λ and α similar to the macroeconomic model Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 14 / 33

The model: special cases K inked SVAR ( λ = α = 0, no shadow rate): Y 1 t = β Y 2 t + B 1 X t + ε 1 t (1) Y 2 t = max { γ Y 1 t + B 2 X t + ε 2 t , b t } , (2) where X t is exogenous and predetermined, ε t iid shocks, ε 1 t ⊥ ⊥ ε 2 t . C ensored SVAR ( λ = 1, α = 0): linear SVAR in ( Y 1 , Y ∗ 2 ) Y 1 t = β Y ∗ 2 t + B 1 X ∗ t + ε 1 t , (3) Y ∗ 2 t = γ Y 1 t + B 2 X ∗ t + ε 2 t , (4) Y 2 t = max { Y ∗ 2 t , b t } , (5) where X ∗ t includes Y ∗ 2 t − j but not Y 2 t − j . Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 15 / 33

Mavroeidis (2019) “Identification at the ZLB” SVARs subject to occasionally binding constraints (CKSVAR) Uses occasionally binding constraints for identification Unconventional policy via “shadow rate” and FG The method can: Identify IRF to monetary policy shocks 1 Obtain bounds on efficacy of unconventional policy 2 Test the “ZLB irrelevance” hypothesis 3 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 16 / 33

The intuition behind identification at the ZLB 4.5 4.0 3.5 3.0 2.5 2.0 1.5 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 17 / 33

The intuition behind identification at the ZLB 4.5 4.0 3.5 3.0 2.5 2.0 1.5 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 17 / 33

The intuition behind identification at the ZLB 4.5 Unconstrained regime 4.0 3.5 3.0 2.5 2.0 1.5 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 17 / 33

The intuition behind identification at the ZLB 4.5 Unconstrained regime 4.0 3.5 3.0 Constrained regime 2.5 2.0 1.5 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 17 / 33

The intuition behind identification at the ZLB 4.5 Unconstrained regime 4.0 3.5 3.0 Constrained regime 2.5 2.0 1.5 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 17 / 33

The intuition behind identification at the ZLB 4.5 Unconstrained regime 4.0 3.5 3.0 Constrained regime 2.5 2.0 1.5 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 17 / 33

Mapping the DSGE model to the CKSVAR In general, no analytical mapping of the DSGE to the CKSVAR No analytical solutions to the DSGE Different interpretation of efficacy of UMP In DSGE, λ ∗ = 0 means no effect of QE, but FG can still be effective In CKSVAR, λ = 0 means no contemporaneous effect of any UMP Perfect mapping when λ = 1 and α = 0 (ZLB irrelevance) Solution to the DSGE: linear SVAR in π t , y t , i ∗ t CKSVAR has high power in detecting deviations from λ = 1 , α = 0 ξ ≡ λ (1 + α ) can be identified, but not separately. Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 18 / 33

Partial identification of ξ (Mavroeidis 2019) Reduced form of CKSVAR for Y 1 t has kink at ZLB: Y 1 t = C 1 X t + C ∗ 12 X ∗ 2 t + u 1 t − � β D t ( C 2 X t + C ∗ 22 X ∗ 2 t + u 2 t − b t ) D t := 1 { Y 2 t = b t } , where β = (1 − ξ ) ( I − ξβγ ) − 1 β, � (1) � 12 − Ω 22 β ′ � � Ω 11 − Ω 12 β ′ � − 1 Ω ′ γ = (2) � β , Ω are identified, but β , γ , ξ are not Identified set consists of all β, γ, ξ that solve (1)-(2) for given � β , Ω Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 19 / 33

Outline Macroeconomic model of UMP 1 Empirical model and identification at the ZLB 2 Testing ZLB irrelevance hypothesis 3 Impact of monetary policy 4 Ikeda (BoJ) Monetary Policy at the ZLB 26 December 2019 19 / 33