Discussion of Does a Big Bazooka Matter? Central Bank Balance-Sheet Policies and Exchange Rates

Discussion of Does a Big Bazooka Matter? Central Bank Balance-Sheet Policies and Exchange Rates by L. Dedola, G. Georgiadis, J. Grab, A. Mehl Ambrogio Cesa-Bianchi (BoE and CfM) ESSIM May 26, 2017 *The views expressed here do not necessarily reflect the position of the Bank of England.

An explosive paper Source: http://www.visualcapitalist.com/ecbs-big-bazookas-tltro-qe/. Great paper, important contribution. Despite the fact that balance sheet policies are the norm, we still know little about how they work. Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 2

What this paper does Assess the impact of (relative) central bank balance-sheet policies on the exchange rate Investigate explicitly channels of transmission Signalling channel, interest rate, differentials and frictions in FX swap markets Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 3

What this paper does Assess the impact of (relative) central bank balance-sheet policies on the exchange rate Investigate explicitly channels of transmission Signalling channel, interest rate, differentials and frictions in FX swap markets Approach: 2SLS 1. Indep. variable Changes in relative CB balance sheets ( BS t+1 + BS t ) 2. Instrument Lagged UMP announcements (a t, a t 1 ) 3. Dep. variable Changes in the nominal exchange rate ( s t ) Interest rate differentials, CIP deviations, currency risk premium Other asset prices, IP, CPI,... Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 3

My discussion Review of 2SLS specification (as I understand it) Three comments 1. First stage 2. Stock vs Flow 3. Role of expectations One suggestion Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 4

The logic of the 2SLS specification Naive approach (mine, not the authors ) Assumption Law of motion of CB balance sheet BS t = ε ump t + ε other t + ρx t 1 Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 5

The logic of the 2SLS specification Naive approach (mine, not the authors ) Assumption Law of motion of CB balance sheet BS t = ε ump t + ε other t + ρx t 1 Instrument Dummy (a t ) that takes value of 1 on announcement days COV[a t, ε ump t ] = 0 COV[a t, ε other t ] = 0 Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 5

The logic of the 2SLS specification Naive approach (mine, not the authors ) Assumption Law of motion of CB balance sheet BS t = ε ump t + ε other t + ρx t 1 Instrument Dummy (a t ) that takes value of 1 on announcement days COV[a t, ε ump t ] = 0 COV[a t, ε other t ] = 0 Estimation Two-stage least squares BS t = b 1 a t + ζ t s t = γ(ˆb 1 a t ) + ξ t Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 5

The logic of the 2SLS specification Naive approach (mine, not the authors ) Assumption Law of motion of CB balance sheet BS t = ε ump t + ε other t + ρx t 1 Instrument Dummy (a t ) that takes value of 1 on announcement days COV[a t, ε ump t ] = 0 COV[a t, ε other t ] = 0 Estimation Two-stage least squares BS t = b 1 a t + ζ t s t = γ(ˆb 1 a t ) + ξ t Issue Reasonable to assume that COV[a t, ε other t ] = 0? Not really... announcement could be endogenous response to ε other t Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 5

The logic of the 2SLS specification Naive approach (mine, not the authors ) Assumption Law of motion of CB balance sheet BS t = ε ump t + ε other t + ρx t 1 Instrument Dummy (a t ) that takes value of 1 on announcement days COV[a t, ε ump t ] = 0 COV[a t, ε other t ] = 0 Estimation Two-stage least squares BS t = b 1 a t + ζ t s t = γ(ˆb 1 a t ) + ξ t Issue Power of instrument? Might be low as actually ε ump t = η t t + η t+1 t Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 6

The logic of the 2SLS specification This paper s (smart) approach Consider balance sheet in t + 1 BS t+1 = η t+1 t+1 + η t+1 t + ε other t+1 + ρx t Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 7

The logic of the 2SLS specification This paper s (smart) approach Consider balance sheet in t + 1 BS t+1 = η t+1 t+1 + η t+1 t + ε other t+1 + ρx t Assume COV[a t, η t+1 t ] = 0 COV[a t, ε other t+1 ] = 0 Then can run BS t+1 = b 1 a t + b 2 X t + ζ t Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 7

The logic of the 2SLS specification This paper s (smart) approach Consider balance sheet in t + 1 BS t+1 = η t+1 t+1 + η t+1 t + ε other t+1 + ρx t Assume COV[a t, η t+1 t ] = 0 COV[a t, ε other t+1 ] = 0 Then can run BS t+1 = b 1 a t + b 2 X t + ζ t Note Role of X t is crucial Controls for ε other t (exclusion restriction) and η t t (endogeneity bias) Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 7

The logic of the 2SLS specification This paper s (smart) approach Consider balance sheet in t + 1 BS t+1 = η t+1 t+1 + η t+1 t + ε other t+1 + ρx t Assume COV[a t, η t+1 t ] = 0 COV[a t, ε other t+1 ] = 0 Then can run BS t+1 = b 1 a t + b 2 X t + ζ t Now can safely estimate second stage s t = γ(ˆb 1 a t ) + ξ Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 8

Comment No.1: First stage equation Baseline first stage equation BS t+1 + BS t = b 1 a t + b 2 X t + b 3 a t 1 + b 4 X t 1 + ζ t Controlling for X t is roughly equivalent to a Cholesky-type of identification Y = VIX t, ESI }{{} t, a t, ( BS t+1 + BS t ), s t X t With financial variables this can be problematic (Gertler and Karadi, 2015) Implicit assumption is that causation goes VIX t /ESI t a t Likely two-way relation between these variables VIX t /ESI t a t Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 9

Comment No.2: Stock vs Flow As authors note, asset prices respond to both contemporaneous and anticipated UMP shocks BS t = ε ump t + ε other t + β 3 X t 1 ε ump t = η t t + η t+1 t +... + η t+h t Define total stock effect announced at time t as BS t+h t = t+h i=0 η t+i t Paper instruments BS t+1 + BS t only How about the remaining BS t+h t ( BS t+1 + BS t )? Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 10

Comment No.2: Stock vs Flow As authors note, asset prices respond to both contemporaneous and anticipated UMP shocks BS t = ε ump t + ε other t + β 3 X t 1 ε ump t = η t t + η t+1 t +... + η t+h t Define total stock effect announced at time t as BS t+h t = t+h i=0 η t+i t Paper instruments BS t+1 + BS t only How about the remaining BS t+h t ( BS t+1 + BS t )? Underestimation of the total effect? Lower power of the instrument? Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 10

Comment No.3: The role of expectations A parallel with the conventional monetary policy literature Consider a short x-minutes window around the announcement on day t at time τ Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 11

Comment No.3: The role of expectations A parallel with the conventional monetary policy literature Consider a short x-minutes window around the announcement on day t at time τ High frequency identification of balance sheet shocks { BSt+h t,τ+x E Z t = τ x [BS t+h t ] on an announcement day 0 otherwise Identification comes from the fact that ε other t can t affect Z t in short window What matters is the surprise relative to expectations! Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 11

Comment No.3: The role of expectations A parallel with the conventional monetary policy literature Consider a short x-minutes window around the announcement on day t at time τ High frequency identification of balance sheet shocks { BSt+h t,τ+x E Z t = τ x [BS t+h t ] on an announcement day 0 otherwise Identification comes from the fact that ε other t can t affect Z t in short window What matters is the surprise relative to expectations! This paper { 1 on an announcement day Z t = 0 otherwise Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 11

Comment No.3: The role of expectations A parallel with the conventional monetary policy literature Consider a short x-minutes window around the announcement on day t at time τ High frequency identification of balance sheet shocks { BSt+h t,τ+x E Z t = τ x [BS t+h t ] on an announcement day 0 otherwise Identification comes from the fact that ε other t can t affect Z t in short window What matters is the surprise relative to expectations! This paper { 1 on an announcement day Z t = 0 otherwise What can go wrong? Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 11

What can go wrong Assume that a negative demand shock hits Outlook for inflation worsens Expectations about UMP, ie E t [BS t+h,t ] > 0 Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 12

What can go wrong Assume that a negative demand shock hits Outlook for inflation worsens Expectations about UMP, ie E t [BS t+h,t ] > 0 On announcement day ECB disappoints expectations BS t+h,t E t [BS t+h,t ] Source: https://www.ft.com/content/3e617908-9d0a-11e5-b45d-4812f209f861. Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 12

What can go wrong Assume that negative demand shock hits Outlook for inflation worsens Expectations about UMP, ie E t [BS t+h,t ] > 0 On announcement day ECB disappoints expectations BS t+h,t E t [BS t+h,t ] Contractionary UMP shock, exchange rate appreciates Downward bias! Source: https://www.bloomberg.com/news/articles/2016-03-10/euro-drops-most-since-november-as-ecb-cuts-rates-expands-qe. Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 13

Suggestion Construct an instrument Z t with high frequency data { BSt+h t,τ+x E Z t = τ x [BS t+h t ] on an announcement day 0 otherwise Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 14

Suggestion Construct an instrument Z t with high frequency data { BSt+h t,τ+x E Z t = τ x [BS t+h t ] on an announcement day 0 otherwise Unfortunately E τ x [BS t+h t ] is not available for much of the sample But virtually any variable for which you have high frequency data would do the trick Long-term government bonds, equity, etc Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 14

Suggestion Construct an instrument Z t with high frequency data { BSt+h t,τ+x E Z t = τ x [BS t+h t ] on an announcement day 0 otherwise Unfortunately E τ x [BS t+h t ] is not available for much of the sample But virtually any variable for which you have high frequency data would do the trick Long-term government bonds, equity, etc Given small number of announcements, could even do it do it narratively... Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 14

Suggestion (cont d) Then, can run following first stage equation BS t+h,t = β 0 + β 1 Z t + ζ t What you would gain More defendable exclusion restriction No need for assumption on direction of causation VIX t /ESI t a t Full account of stock effect No downward bias due to expectations Richer time variation in the instrument Instrument proportional to size of the surprise Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 15

(Very) minor issues Hard to follow the derivation of the 2SLS specification Plot CIP deviations and λ t Puzzling that a FED t is weak instrument Notation: sometimes you use BS, BS, or B Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 16

Summing up Great paper Provides new insights on (unconventional) monetary policy transmission Very much welcome, need more of this type of papers Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 17

Summing up Great paper Provides new insights on (unconventional) monetary policy transmission Very much welcome, need more of this type of papers Somewhat puzzlingly small and/or insignificant impact of balance sheet policies Underestimation the bazooka effect? Expectations + Stock effect (downward bias) Cholesky (unknown) Results could be even sharper Discussion of Dedola, Georgiadis, Grab, Mehl: Does a Big Bazooka Matter? 17