Multivariate Time

Series

Firmin Doko
Tchatoka

Introduction Identification of SVARs Using Sign Restrictions and

Review: Zero
contemporaneous
Proxy-SVAR Methods
and long-run
restrictions

Sign restrictions

External Instruments
Firmin Doko Tchatoka
(or Proxy SVARs)

Combining Sign
Restrictions and
The University of Adelaide
External Instruments

Practical Examples [email protected]

END
 Course outline
Multivariate Time
Series

Firmin Doko
Tchatoka
1 Introduction

Introduction
2 Review: Zero contemporaneous and long-run restrictions
Review: Zero
contemporaneous
and long-run
restrictions 3 Sign restrictions
Sign restrictions

External Instruments 4 External Instruments (or Proxy SVARs)

(or Proxy SVARs)

Combining Sign
Restrictions and 5 Combining Sign Restrictions and External Instruments
External Instruments

Practical Examples
6 Practical Examples
END

7 END
 Multivariate Time
Series

Firmin Doko
Tchatoka

Introduction

Review: Zero
contemporaneous
and long-run
restrictions

Sign restrictions
Introduction
External Instruments
(or Proxy SVARs)

Combining Sign
Restrictions and
External Instruments

Practical Examples

END
 Common SVAR Identification Schemes
Multivariate Time
Series

Firmin Doko
Tchatoka ◦ Zero (recursive) contemporaneous restrictions
Introduction

Review: Zero
◦ Zero (recursive) long-run restrictions
contemporaneous
and long-run
restrictions ◦ Sign restrictions
Sign restrictions

External Instruments
(or Proxy SVARs) ◦ External instruments
Combining Sign
Restrictions and
External Instruments ◦ Combining sign restrictions and external instruments
Practical Examples

END ◦ Other (narrative sign restrictions, maximization of forecast error variance,

...)
 Multivariate Time
Series

Firmin Doko
Tchatoka

Introduction

Review: Zero
contemporaneous
and long-run
restrictions

Sign restrictions
Review: Zero contemporaneous and long-run
External Instruments
(or Proxy SVARs)
restrictions
Combining Sign
Restrictions and
External Instruments

Practical Examples

END
 Zero contemporaneous restrictions
Multivariate Time
Series
• Intuition. Identification is achieved by assuming that some shocks have
Firmin Doko zero contemporaneous effect on some of the endogenous variables
Tchatoka

Introduction
• References: Sims (1980), Christiano, Eichenbaum, Evans (1999)
Review: Zero
contemporaneous
• For example, assume that monetary policy works with a lag and has no
and long-run
restrictions
contemporaneous effects on output
Sign restrictions • How can restrictions on the effect of structural shocks be imposed?
External Instruments − Solution: zero restrictions on the impact matrix
(or Proxy SVARs)

Combining Sign → Coefficient capturing the contemporaneous effect of monetary policy on

Restrictions and
External Instruments output growth is set to zero:
Practical Examples         Demand 
GDPt φ11 φ12 GDPt−1 b11 0 εt
END = + (1)
rt φ21 φ22 rt−1 b21 b22 εMonPol
t

• Implication: 3 structural parameters to estimate (instead of 4) and 3

restrictions implied by Σu = E(ut0 ut ), ut = Bεt .
 Zero long-run restrictions
Multivariate Time
Series
• Intuition. Identification is achieved by assuming that some shocks have
Firmin Doko zero cumulative effect on some of the endogenous variables in the long-run
Tchatoka
• References: Blanchard and Quah (1989), Galı̀ (1999)
Introduction

Review: Zero
contemporaneous
• For example, assume that monetary policy is neutral in the long-run and
and long-run
restrictions
has no cumulative effect on the level of output
Sign restrictions • How can restrictions on the long-run cumulative effect of structural
External Instruments shocks be imposed?
(or Proxy SVARs)
− Re-write the VAR as (where Xt = (GDPt , rt )0 ):
Combining Sign
Restrictions and
External Instruments
Xt = ΦXt−1 + Bεt (2)
Practical Examples

END → If a shock εt hits in t, its cumulative impact on Xt in the long-run is:

Xt,t+∞ = Bε + ΦBεt + Φ2 Bεt + · · · + Φ∞ Bεt
|{z}t | {z } | {z }
impact at t impact at t+1 etc. . .
−1
≡ (I2 − Φ) Bεt = C εt , C = (I2 − Φ)−1 B (3)
 Zero long-run restrictions (cont’d)
Multivariate Time
Series • Cumulative impact:
Firmin Doko
Tchatoka Xt,t+∞ = (I2 − Φ)−1 Bεt = C εt
Introduction

Review: Zero
contemporaneous − For output growth, GDPt,t+∞ is the effect of εt on the level of output
and long-run
restrictions
− C = (I2 − Φ)−1 B captures the cumulative effect of εt on Xt from t to ∞.
Sign restrictions

External Instruments • Go back to our output growth/policy rate example:

(or Proxy SVARs)
     Demand 
Combining Sign GDPt,t+∞ c11 c12 εt
Restrictions and = (4)
External Instruments rt,t+∞ c21 c22 εMonPol
t
Practical Examples

END
• First equation is: GDPt,t+∞ = c11 εDemand
t + c12 εMonPol
t
− c12 ≡ impact of a monetary policy shock (hitting in t) on the level of GDP
in the long-run
− Long-run neutrality of monetary policy means that c12 = 0
 Multivariate Time
Series

Firmin Doko
Tchatoka

Introduction

Review: Zero
contemporaneous
and long-run
restrictions

Sign restrictions
Sign restrictions
External Instruments
(or Proxy SVARs)

Combining Sign
Restrictions and
External Instruments

Practical Examples

END
 Sign restrictions
Multivariate Time
Series
◦ Intuition. Exploit prior beliefs(typically informed by theoretical models)
Firmin Doko about the sign that certain shocks should have on certain endogenous
Tchatoka
variables
Introduction

Review: Zero
◦ References: Faust (1998), Canova and De Nicolo (2002), Uhlig (2005),
contemporaneous Pegan and Ouilli (2016)
and long-run
restrictions

Sign restrictions
◦ For example
External Instruments
(or Proxy SVARs)
− Demand shocks should lead to an increase in output and interest rates
Combining Sign − Monetary policy shocks should lead to a fall in output and an increase in
Restrictions and
External Instruments interest rates
Practical Examples Demand Monetary Policy
END (εDemand
t ) (εMonPol
t )
Output Growth (gpdt ) + -
Short-rate Int. Rate (rt ) + +

Table: Sign restriction

 Sign restrictions (cont’d)
Multivariate Time
Series ◦ How can we impose restrictions on the signs of the effect of a
Firmin Doko structural shock?
Tchatoka

Introduction
◦ Key intuition is based on the following three steps
Review: Zero
contemporaneous
1 Consider a random orthonormal matrix Q such that QQ 0 = I2
and long-run
restrictions 2 Consider the lower triangular B matrix corresponding to the Cholesky factor
Sign restrictions of Σu = E(ut0 ut ) [where ut = Bεt ]: Σu = PP 0
External Instruments
(or Proxy SVARs) 3 The following equality holds:
Combining Sign
Restrictions and Σu = PP 0 = PQQ 0 P 0 = (PQ) (PQ)0 (5)
External Instruments | {z } | {z }
Practical Examples B B0

END
◦ The matrix B = PQ is a valid ‘candidate’ impact matrix that solves the
identification problem!
− Differently from P, the matrix PQ is not lower triangular anymore.
 Sign restrictions– Orthonormal matrix
Multivariate Time
Series

Firmin Doko
Tchatoka
◦ An orthonormal matrix Q is a real square matrix whose columns and
rows are orthogonal unit vectors
Introduction

Review: Zero ◦ It means that if you take for example two 2 × 1 vectors q1 and q2 , then
contemporaneous
and long-run the matrix Q = (q1, q2) is orthonormal if
restrictions

Sign restrictions 1 Consider a random orthonormal matrix Q such that QQ 0 = I2

External Instruments
(or Proxy SVARs) ? The vectors have unit norm: kqi k = 1 for all i = 1, 2
Combining Sign
Restrictions and ? The vectors are mutually orthogonal: q10 q2 = 0
External Instruments

Practical Examples ◦ It follows that QQ 0 = I and Q 0 = Q −1

END
◦ Note. You can draw in Matlab/Stata infinite matrices that satisfy the
above conditions
 Sign restrictions– How to achieve identification?
Multivariate Time
Series
◦ Q is a random matrix. How can we check that B = PQ represents a
Firmin Doko plausible solution?
Tchatoka
◦ Solution. Check that the effects of shocks implied by B = PQ satisfy a
Introduction

Review: Zero
set of a priori sign restrictions. That is:
contemporaneous
and long-run 1 Consider the structural representation of our VAR
restrictions         Demand 
Sign restrictions GDPt φ11 φ12 GDPt−1 b11 b12 εt
= + (6)
External Instruments rt φ21 φ22 rt−1 b21 b22 εMonPol
t
(or Proxy SVARs)

Combining Sign 2 Then check that the elements of B satisfy

Restrictions and
External Instruments
Demand Monetary Policy
Practical Examples
(εDemand
t ) (εMonPol
t )
END
Output Growth (gpdt ) b11 > 0? b12 < 0?
Short-rate Int. Rate (rt ) b21 > 0? b22 > 0?

Table: Sign restriction

 Sign restrictions in steps
Multivariate Time
Series
◦ Perform N replications of the following steps
Firmin Doko
Tchatoka
(1) Draw a random orthonormal matrix Q
Introduction

Review: Zero
(2) Compute B = PQ where P is the Cholesky decomposition of the reduced
contemporaneous form residuals covariance matrix Σu
and long-run
restrictions
(3) Compute the impact effects of shocks associated with B
Sign restrictions

External Instruments (4) Are the sign restrictions satisfied?

(or Proxy SVARs)

Combining Sign (4.1) Yes. Store B and go back to (1).

Restrictions and
External Instruments
(4.2) No. Discard B and go back to (1).
Practical Examples

END ◦ All matrices in the set B (i) (for i = 1, 2, . . . , N) represent admissible

solutions to the identification problem
◦ In this sense, sign restricted VARs are only set identified.
 Multivariate Time
Series

Firmin Doko
Tchatoka

Introduction

Review: Zero
contemporaneous
and long-run
restrictions

Sign restrictions
External Instruments (or Proxy SVARs)
External Instruments
(or Proxy SVARs)

Combining Sign
Restrictions and
External Instruments

Practical Examples

END
 External instruments
Multivariate Time
Series ◦ Intuition. Exploit information from a variable that is external to the VAR,
Firmin Doko
Tchatoka
but that is correlated with a particular shock of interest and uncorrelated
with other shocks (the instrument)
Introduction

Review: Zero ◦ References: Stock and Watson (2012), Mertens and Ravn (2013), Gertler
contemporaneous
and long-run and Karadi (2015), Mertens and Olea (2018), Olea-Stock-Watson (2020)
restrictions

Sign restrictions ◦ For example:

External Instruments
(or Proxy SVARs) − assume that you have some ‘narrative’ series of policy surprises (i.e., that
Combining Sign are not just a response of the central bank to some development in the
Restrictions and
External Instruments economy)
Practical Examples
◦ How can this help in finding the B matrix?
END

◦ Key element. If an instrument that is correlated with a shock of

interest and uncorrelated with all other shocks can be found, we can
implement an IV method as explained in the next slide
 External instruments (cont’d)
Multivariate Time
Series

Firmin Doko ◦ Go back to our output growth/policy rate example

Tchatoka

Introduction − Assume that such an instrument exists (Zt ) for the monetary policy shock
Review: Zero with the following properties:
contemporaneous
and long-run
E εDemand Zt0 = 0
 
restrictions t

E εMonPol Zt0 = c 6= 0
 
Sign restrictions
t (7)
External Instruments
(or Proxy SVARs)

Combining Sign
Restrictions and
− Then, we can identify one column (in this example, the second one) of the
External Instruments B matrix:
Practical Examples         Demand 
GDPt φ11 φ12 GDPt−1 · b12 εt
END = + (8)
rt φ21 φ22 rt−1 · b22 εMonPol
t
 How external instruments identification work?
Multivariate Time
Series
◦ Recall that the reduced form residuals are a linear combination of the
Firmin Doko two structural shocks:
Tchatoka (
ugdp,t = b11 εDemand
t + b12 εMonPol
t
Introduction (9)
ur ,t = b21 εDemand
t + b ε
22 t
MonPol
Review: Zero
contemporaneous
and long-run ◦ The OLS estimate of β in the following ‘first-stage’ regression (scalar Zt
restrictions
case) identifies b22 up to a scaling factor:
Sign restrictions

External Instruments urt = βZt + ξt (10)

(or Proxy SVARs)

Combining Sign
◦ To see that, recall that β can be written as β = cov (urt , Zt )/var (Zt ):
Restrictions and − Focus on the Cov term and plug in the definition of urt to get:
External Instruments

Practical Examples Cov (urt , Zt ) = Cov (21 εDemand

t + b22 εMonPol
t , Zt ) = b22 Cov (εMonPol
t , Zt ) = b22 c
END

− It follows that β = b22 c/var (Zt )

◦ As c is an unknown constant, b22 = βVar (Zt )/c is only identified to a

scaling factor
 How external instruments identification work (cont’d)?
Multivariate Time
Series ◦ The OLS estimate of γ in the following ‘second-stage’ regression identifies
Firmin Doko the ratio b12 /b22 :
Tchatoka
 b c 
Introduction 22
udgp,t = γ ûrt + ζt = Z t + ζt (11)
Review: Zero var (Zt )
contemporaneous
and long-run
restrictions ◦ To see that, recall that β can be written as γ = cov (ugdp,t , ûrt )/var (ûrt ):
Sign restrictions − Focus on the Cov term and plug in the definition of ugdp,t and ûrt to get:
External Instruments
(or Proxy SVARs)  b22 c  b12 b22 c 2
Combining Sign Cov (urt , Zt ) = Cov b11 εDemand
t + b12 εMonPol
t , = (12)
Restrictions and var (Zt ) var (Zt )
External Instruments

Practical Examples − Then focus on the Var term to get:

END  b c
22
 b2 c 2
var (ûrt ) = var Zt = 22 (13)
var (Zt ) var (Zt )
b12
◦ It follows that γ = b22 .
 External instruments– Partial identification
Multivariate Time
Series

Firmin Doko
Tchatoka ◦ We can normalize the effect of εMonPol
t on rt to 1, i.e., b22 = 1
Introduction

Review: Zero
◦ And quantify the effect of εMonPol
t on GDPt as b12 = γ
contemporaneous
and long-run
restrictions ◦ In other words, we have identified the column of the B matrix of the
Sign restrictions
structural VAR representation up to a scaling factor:
External Instruments
(or Proxy SVARs)         Demand 
GDPt φ11 φ12 GDPt−1 · γ εt
Combining Sign = + (14)
Restrictions and rt φ21 φ22 rt−1 · 1 εMonPol
t
External Instruments

Practical Examples

END
◦ Note. It is actually possible to work out the true values of B; see
footnote 4 of Gertler and Karadi (2015).
 Multivariate Time
Series

Firmin Doko
Tchatoka

Introduction

Review: Zero
contemporaneous
and long-run
restrictions

Sign restrictions
Combining Sign Restrictions and External
External Instruments
(or Proxy SVARs)
Instruments
Combining Sign
Restrictions and
External Instruments

Practical Examples

END
 Sign restrictions + External instruments
Multivariate Time
Series

Firmin Doko
Tchatoka ◦ Intuition. Identify one (or more) columns of B with external instruments
and conditional on that the remaining columns with sign restrictions
Introduction

Review: Zero
contemporaneous
and long-run
◦ References: Cesa-Bianchi and Sokol (2019), Cesa-Bianchi and Ferrero
restrictions (2020)
Sign restrictions

External Instruments ◦ For example:

(or Proxy SVARs)

Combining Sign
Restrictions and − assume that there are two shocks that imply similar signs (so that sign
External Instruments
restrictions are not enough to identify the shocks), but you have an
Practical Examples
instrument for one of the two shocks
END

◦ How can we find the B matrix in this case?

 Sign restrictions + External instruments (cont’d)
Multivariate Time
Series

Firmin Doko
Tchatoka ◦ Consider a n-variable version of our simple structural VAR(1):
y1t φ11 φ12 ... φ1n y1,t−1 b11 b12 ... b1n ε1t
Introduction
       

Review: Zero  y2t   φ21 φ22 ... φ2n   y2,t−1   b21 b22 ... b2n   ε2t 
 . = . .. .. .. + . .. ..  .
       
contemporaneous
 ..   .. .. 
  .. ..   ..

and long-run . . .  . . . . 
restrictions
ynt φn1 φn2 ... φnn yn,t−1 bn1 bn2 ... bnn εnt
Sign restrictions

External Instruments
(or Proxy SVARs) ◦ Assume that
Combining Sign
Restrictions and
External Instruments − The first structural shock (ε1t ) can be identified with an external instrument
Practical Examples

END − The remaining structural shocks (ε2t , . . . , εnt ) can be identified with sign
restrictions
 Sign restrictions + External instruments (cont’d)
Multivariate Time
Series
◦ Partition the structural matrix B as [b B]:
Firmin Doko
Tchatoka
b B
Introduction b
11 b12 ... b1n 
Review: Zero
contemporaneous b21 b22 ... b2n 
and long-run B = . .. .. .. 
restrictions  .. . . . 
Sign restrictions
bn1 bn2 ... bnn
External Instruments
(or Proxy SVARs)

Combining Sign
Restrictions and
External Instruments
− Column vector b captures the impact of the first shock, matrix B captures
Practical Examples
the impact of the remaining shocks
END − We have seen above how to identify b with external instruments
− Now, the question is: Once b is known, how can we find a B matrix
that satisfies a set of sign restrictions?
 Sign restrictions + External instruments (cont’d)
Multivariate Time
Series

Firmin Doko ◦ Let C be the Cholesky decomposition of Σu (reduced-form VAR error

Tchatoka
covariance matrix). Find a normal vector q of dimension n × 1 that
Introduction rotates the first column of C into the vector b, so that Cq = b
Review: Zero
contemporaneous
and long-run
◦ Given q, build a n × (n − 1) matrix Q such that Q = [q Q] is
restrictions orthonormal, i.e., [q Q][q Q]0 = QQ 0 = I
Sign restrictions

External Instruments ◦ As Q is an orthonormal matrix we have:

(or Proxy SVARs)
Σu = CC 0 = CQQ 0 C 0 = (CQ)(CQ)0
Combining Sign
Restrictions and
External Instruments ◦ So B = CQ is a valid candidate matrix that solves the identification
Practical Examples problem as
END
− Σu = (CQ)(CQ)0 holds
− The first column of CQ is b
 Sign restrictions + External instruments– steps
Multivariate Time
Series

Firmin Doko
(1) Identify b, the first column of B = [q B], with the external instrument
Tchatoka
(2) Compute the Cholesky decomposition C of the reduced form residuals
Introduction
covariance matrix Σu
Review: Zero
contemporaneous
and long-run (3) Find a normal vector q that rotates the first column of C into the vector
restrictions
b, namely Cq = b
Sign restrictions

External Instruments (3.i) Given q, build the remaining n − 1 columns of an orthonormal matrix
(or Proxy SVARs)
Q = [q Q]
Combining Sign
Restrictions and
External Instruments (3.ii) The matrix CQ then represents a candidate identification scheme because:
Practical Examples (CQ)(CQ)0 = Σu and C [q Q] = [b B]
END
(3.iii) If B] satisfies the sign restrictions, retain it. Otherwise, go back to (3.i)

(4) Go back to (1) and repeat N times.

 Multivariate Time
Series

Firmin Doko
Tchatoka

Introduction

Review: Zero
contemporaneous
and long-run
restrictions

Sign restrictions
Practical Examples
External Instruments
(or Proxy SVARs)

Combining Sign
Restrictions and
External Instruments

Practical Examples

END
 The VAR Toolbox
Multivariate Time
Series
◦ Illustrate how VARs work through a set of examples using the VAR
Firmin Doko
Tchatoka Toolbox 3.0
Introduction
◦ The VAR Toolbox is a collection of Matlab routines to perform VAR
Review: Zero
contemporaneous
analysis
and long-run
restrictions
− Codes are available at: https://github.com/ambropo/VAR-Toolbox
Sign restrictions

External Instruments − No installation is required. Simply clone the folder from Github and add the
(or Proxy SVARs)
folder (with sub-folders) to your Matlab path
Combining Sign
Restrictions and
External Instruments
− To save figures in high quality format, you need to download an install
Ghostscript (freely available at www.ghostscript.com).
Practical Examples

END ? The first time you’ll be saving a figure using the Toolbox, you’ll be asked to
locate Ghostscript on your local drive

◦ Do some replication exercises with a few well-known papers.

 The VAR Toolbox– Notes
Multivariate Time
Series

Firmin Doko
Tchatoka
◦ To avoid clashes with functions from other toolboxes, it is recommendable
Introduction to add and remove the Toolbox at beginning and end of your scripts
Review: Zero
contemporaneous
and long-run ◦ If you download the toolbox to /User/VAR-Toolbox/, you can simply add
restrictions
the following lines at the beginning and end of your script:
Sign restrictions

External Instruments
(or Proxy SVARs)
addpath(genpath(’/User/VAR-Toolbox/v3dot0/’))
Combining Sign
Restrictions and
External Instruments
...
Practical Examples

END
rmpath(genpath(’/User/VAR-Toolbox/v3dot0’)).
 Example 1: Stock&Watson – Zero short-run restrictions
Multivariate Time
Series
◦ Stock and Watson (2001). “Vector Autoregressions,” Journal of
Firmin Doko Economic Perspectives
Tchatoka
◦ US quarterly data from 1960:Q1 to 2000:Q4
Introduction ◦ Objective. Infer the causal influence of monetary policy on
Review: Zero
contemporaneous
unemployment and inflation
and long-run
restrictions
◦ Assume a VAR with p = 4 with inflation (πt ), unemployment (ut ), and
Sign restrictions the fed funds rate (rt )
External Instruments ◦ Key identifying assumptions
(or Proxy SVARs)
− MP (rt ) reacts contemporaneously to movements in inflation and in
Combining Sign
Restrictions and unemployment
External Instruments − MP shocks (εMonPol
t ) do not affect inflation and unemployment within the
Practical Examples quarter of the shock:
END
   (k) (k) (k)
    
πt X4 φ11 φ12 φ13 πt−k b11 0 0 ε1t
 (k) (k) (k)
 ut  = φ23   ut−k  +  b21 b22 0  ε2t

 φ21 φ22 
rt k=1 (k)
φ31
(k)
φ32
(k)
φ33 rt−k b31 b32 b33 εMonPol
t
 SW (2001) with the VAR Toolbox
Multivariate Time
Series ◦ In Matlab, set lag length to 4 and estimate a VAR with a constant:
Firmin Doko % Set up and estimate VAR
Tchatoka
det = 1;
Introduction nlags = 4;
Review: Zero
contemporaneous
[VAR, VARopt] = VARmodel(X,nlags,det);
and long-run
restrictions
◦ Then set the option for recursive identification VARopt.ident =’short’ and
Sign restrictions
compute the IRs with the VARir function:
External Instruments
(or Proxy SVARs)
− Note that the ordering of the variables matter!
Combining Sign
Restrictions and % For zero contemporaneous restrictions set:
External Instruments VARopt.ident = ’short’;
Practical Examples % Compute IR
END [IR, VAR] = VARir(VAR,VARopt);
◦ Note that the second output of the VARir function is VAR again:
− This is because the VAR structure is updated with the B matrix
corresponding to the identification scheme chosen
 SW (2001) with the VAR Toolbox (cont’d)
Multivariate Time
Series

Firmin Doko ◦ The VARirband function allows to compute confidence intervals:

Tchatoka

Introduction
% Compute IR
Review: Zero
[IR, VAR] = VARir(VAR,VARopt);
contemporaneous
and long-run
restrictions
◦ You can control the options of the bootstrap procedure by modifying the
Sign restrictions
VARopt structure (before running VARir )
External Instruments
(or Proxy SVARs) ◦ For example
Combining Sign
Restrictions and % Some options for the bootstrap
External Instruments
VARopt.ndraws = 1000; % Number of draws
Practical Examples
VARopt.pctg = 95; % Level for confidence intervals
END
VARopt.method = ’bs’; % ’bs’ sampling with replacement; ’wild’ wild
bootstrap
 Example 2: Blanchard&Quah– Zero long-run restrictions
Multivariate Time
Series
◦ Blanchard and Quah (1989). “The Dynamic Effects of Aggregate
Firmin Doko Demand and Supply Disturbances”, American Economic Review
Tchatoka
◦ US quarterly data from 1948:Q1 to 1987:Q4
Introduction

Review: Zero
◦ Objective. Identify the effects of demand and supply shocks on output
contemporaneous and unemployment
and long-run
restrictions
◦ Assume a bivariate VAR with p = 8 with with output growth (yt ) and
Sign restrictions
unemployment (ut )
External Instruments
(or Proxy SVARs) ◦ Key identifying assumptions. Demand-side shocks have no long-run
Combining Sign
Restrictions and
effect on the level of output, while supply-side shocks do
External Instruments
◦ Blanchard and Quah impose zero long-run restrictions on the cumulative
Practical Examples

END
effect of demand shocks on output growth (i.e. on output level) to
identify the shocks:
     Supply 
yt,t+∞ c11 0 εt
=+
utt+∞ c21 c22 εDemand
t
 BQ (1989) with the VAR Toolbox
Multivariate Time
Series

Firmin Doko
◦ In Matlab, set lag length to 8 and estimate a VAR with a constant:
Tchatoka % Set up and estimate VAR
Introduction det = 1;
Review: Zero nlags = 8;
contemporaneous
and long-run [VAR, VARopt] = VARmodel(X,nlags,det);
restrictions

Sign restrictions ◦ Then set the option for zero long-run restrictions VARopt.ident =’long’
External Instruments and compute the IRs with the VARir function:
(or Proxy SVARs)

Combining Sign
Restrictions and
− Note that the ordering of the variables matter!
External Instruments % For zero contemporaneous restrictions set:
Practical Examples VARopt.ident = ’long’;
END % Compute IR
[IR, VAR] = VARir(VAR,VARopt);

◦ The B matrix implied by the zero long-run restrictions is stored in VAR.B

 Example 3: Uhlig (2005)– Sign restrictions
Multivariate Time
Series
◦ Uhlig (2005). “What are the effects of monetary policy on output?
Firmin Doko
Tchatoka Results from an agnostic identification procedure,” JME
Introduction ◦ US monthly data from 1965:M1 to 2003:M12
Review: Zero
contemporaneous ◦ Objective. Infer the causal effect of monetary policy on real GDP
and long-run
restrictions
◦ Assume a VAR with p = 12 with real GDP, real GDP deflator, a
Sign restrictions

External Instruments
commodity price index, total reserves, non-borrowed reserves, and the fed
(or Proxy SVARs) funds rate
Combining Sign
Restrictions and
External Instruments
◦ Key identifying assumptions. According to conventional wisdom,
Practical Examples
monetary contractions should
END − Raise the federal funds rate
− Lower prices
− Decrease non-borrowed reserves
− Real GDP is left unrestricted
 Example 3: Uhlig (2005)– Sign restrictions (cont’d)
Multivariate Time
Series

Firmin Doko ◦ Uhlig (2005) imposes the following sign restrictions on the impulse
Tchatoka
responses of the VAR
Introduction

Review: Zero
Monetary Policy shock
contemporaneous
and long-run
Real GDP ?
restrictions Real GDP deflator <0
Sign restrictions Commodity price index ?
External Instruments
(or Proxy SVARs)
Total reserves ?
Combining Sign Non-borrowed reserves <0
Restrictions and
External Instruments
Fed Funds Rate >0
Practical Examples
Table: Uhlig (2005) sign restriction
END

◦ Restrictions are imposed for 6 periods

 Uhlig (2005) with the VAR Toolbox
Multivariate Time
Series
◦ In Matlab, the sign restrictions can be set as follows:
Firmin Doko % Define the shock names
Tchatoka
VARopt.snames = ’Mon. Policy Shock’;
Introduction % Define sign restrictions: positive 1, negative -1, unrestricted 0
Review: Zero SIGN = [ 0,0,0,0,0,0; % Real GDP
contemporaneous
and long-run -1,0,0,0,0,0; % Deflator
restrictions

Sign restrictions
-1,0,0,0,0,0; % Commodity Price
External Instruments
0,0,0,0,0,0; % Total Reserves
(or Proxy SVARs) -1,0,0,0,0,0; % NonBorr. Reserves
Combining Sign
Restrictions and
1,0,0,0,0,0]; % Fed Funds
External Instruments % Define the number of steps the restrictions are imposed for:
Practical Examples
VARopt.sr−hor = 6;
END
◦ Sign restriction routine is then implemented with the SR function
% Function SR performs the sign restrictions identification and computes
% IRs, VDs, and HDs. All the results are stored in SRout
SRout = SR(VAR,SIGN,VARopt);
 Example 4: Gertler and Karadi (2015)– External
instruments
Multivariate Time
Series
◦ Gertler and Karadi (2015). “Monetary Policy Surprises, Credit Costs, and
Firmin Doko Economic Activity,” American Economic Journal: Macroeconomics
Tchatoka
◦ US monthly data US monthly data from 1979:M7 to 2012:M6
Introduction

Review: Zero
contemporaneous
◦ Objective. Infer the causal influence of monetary policy on real GDP
and long-run
restrictions ◦ Assume a VAR with p = 12 with industrial production, the consumer price
Sign restrictions index, the 1-year T-bill interest rate, and the Excess Bond Premium
External Instruments
(or Proxy SVARs)
◦ Key identifying assumptions. There exists an external instrument (Zt )
Combining Sign
Restrictions and such that
External Instruments
E εit Zt0 = 0 for i 6= MonPol
 
Practical Examples

E εMonPol Zt0 = c 6= 0
 
END
t

◦ That is: Zt is correlated with the monetary policy shock and uncorrelated
with all other structural shocks in the system
 GK (2015) instruments– High frequency monetary
policy surprises
Multivariate Time
Series

Firmin Doko
◦ Ingredients
Tchatoka
− Intra-daily data (τ denotes minutes)
Introduction

Review: Zero − A monetary policy announcement on day t at time τ (e.g., FOMC decision)
contemporaneous
and long-run
restrictions
− A policy indicator r (e.g., fed funds target)
Sign restrictions j  
− Price of futures contract on r for j days ahead Pt,τ = 100 − Et,τ r j
External Instruments
(or Proxy SVARs)
◦ Monetary policy surprise
Combining Sign
Restrictions and
j j j  j  j
External Instruments
St,τ = −(Pt,τ +20 − Pt,τ −10 ) = Et,τ +20 r − Et,τ −10 r
Practical Examples

END

◦ Intuition. Only monetary policy shocks affect the futures prices in this
short 30-minute window!
 External instruments with the VAR Toolbox
Multivariate Time
Series
◦ In Matlab, first add the instrument to the VAR structure:
Firmin Doko % Identification is achieved with the external instrument, which needs
Tchatoka
% to be added to the VAR structure
Introduction
VAR.IV = IV;
Review: Zero
contemporaneous ◦ Then update options for identification and computation of error bands
and long-run
restrictions
% Update the options in VARopt to be used in IR calculations and plots
Sign restrictions
VARopt.ident = ’iv’;
External Instruments
(or Proxy SVARs) VARopt.method = ’wild’;
Combining Sign
Restrictions and
◦ Finally, compute the IR with the VARir function
External Instruments
− The code instruments the residual of the first equation, so the ordering of
Practical Examples
the variables matter!
END
% Compute IR
[IR, VAR] = VARir(VAR,VARopt);
◦ The b matrix implied by the external instrument is stored in VAR.b
and additional info on the first-stage is stored in in VAR.FirstStage
 Multivariate Time
Series

Firmin Doko
Tchatoka

Introduction

Review: Zero
contemporaneous
and long-run
restrictions

Sign restrictions
END
External Instruments
(or Proxy SVARs)

Combining Sign
Restrictions and
External Instruments

Practical Examples

END