Welcome to Scribd. Sign in or start your free trial to enjoy unlimited e-books, audiobooks & documents.Find out more
Download
Standard view
Full view
of .
Look up keyword or section
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
Var Models

Var Models

Ratings: (0)|Views: 4|Likes:
Published by Aderaw Gashayie

More info:

Published by: Aderaw Gashayie on Jul 24, 2012
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

02/18/2014

pdf

text

original

 
This is page 383Printer: Opaque this
11
Vector Autoregressive Models forMultivariate Time Series
11.1 Introduction
The
vector autoregression 
(VAR)
model 
is one of the most successful,
exi-ble, and easy to use models for the analysis of multivariate time series. It isa natural extension of the univariate autoregressive model to dynamic mul-tivariate time series. The VAR model has proven to be especially useful fordescribing the dynamic behavior of economic and
nancial time series andfor forecasting. It often provides superior forecasts to those from univari-ate time series models and elaborate theory-based simultaneous equationsmodels. Forecasts from VAR models are quite
exible because they can bemade conditional on the potential future paths of speci
ed variables in themodel.In addition to data description and forecasting, the VAR model is alsoused for structural inference and policy analysis. In structural analysis, cer-tain assumptions about the causal structure of the data under investiga-tion are imposed, and the resulting causal impacts of unexpected shocks orinnovations to speci
ed variables on the variables in the model are summa-rized. These causal impacts are usually summarized with impulse responsefunctions and forecast error variance decompositions.This chapter focuses on the analysis of covariance stationary multivari-ate time series using VAR models. The following chapter describes theanalysis of nonstationary multivariate time series using VAR models thatincorporate cointegration relationships.
 
384 11. Vector Autoregressive Models for Multivariate Time Series
This chapter is organized as follows. Section 11.2 describes speci
cation,estimation and inference in VAR models and introduces the
S+FinMetrics
function
VAR
. Section 11.3 covers forecasting from VAR model. The discus-sion covers traditional forecasting algorithms as well as simulation-basedforecasting algorithms that can impose certain types of conditioning infor-mation. Section 11.4 summarizes the types of structural analysis typicallyperformed using VAR models. These analyses include Granger-causalitytests, the computation of impulse response functions, and forecast errorvariance decompositions. Section 11.5 gives an extended example of VARmodeling. The chapter concludes with a brief discussion of Bayesian VARmodels.This chapter provides a relatively non-technical survey of VAR models.VAR models in economics were made popular by Sims (1980). The de
nitivetechnical reference for VAR models is L¨utkepohl (1991), and updated sur-veys of VAR techniques are given in Watson (1994) and L¨utkepohl (1999)and Waggoner and Zha (1999). Applications of VAR models to
nancialdata are given in Hamilton (1994), Campbell, Lo and MacKinlay (1997),Cuthbertson (1996), Mills (1999) and Tsay (2001).
11.2 The Stationary Vector Autoregression Model
Let
Y
t
= (
y
1
t
,y
2
t
,...,y
nt
)
0
denote an (
n
×
1) vector of time series variables.The basic
p
-lag
vector autoregressive 
(VAR(
 p
)) model has the form
Y
t
=
c
+
Π
1
Y
t
1
+
Π
2
Y
t
2
+
· · ·
+
Π
 p
Y
t
 p
+
ε
t
, t
= 1
,...,T 
(11.1)where
Π
i
are (
n
×
n
) coe
cient matrices and
ε
t
is an (
n
×
1) unobservablezero mean white noise vector process (serially uncorrelated or independent)with time invariant covariance matrix
Σ
. For example, a bivariate VAR(2)model equation by equation has the form
µ
y
1
t
y
2
t
=
µ
c
1
c
2
+
µ
π
111
π
112
π
121
π
122
µ
y
1
t
1
y
2
t
1
(11.2)+
µ
π
211
π
212
π
221
π
222
µ
y
1
t
2
y
2
t
2
+
µ
ε
1
t
ε
2
t
(11.3)or
y
1
t
=
c
1
+
π
111
y
1
t
1
+
π
112
y
2
t
1
+
π
211
y
1
t
2
+
π
212
y
2
t
2
+
ε
1
t
y
2
t
=
c
2
+
π
121
y
1
t
1
+
π
122
y
2
t
1
+
π
221
y
1
t
1
+
π
222
y
2
t
1
+
ε
2
t
where
cov
(
ε
1
t
,ε
2
s
) =
σ
12
for
t
=
s
; 0 otherwise. Notice that each equationhas the same regressors — lagged values of 
y
1
t
and
y
2
t
. Hence, the VAR(
 p
)model is just a
seemingly unrelated regression 
(SUR) model with laggedvariables and deterministic terms as common regressors.
 
11.2 The Stationary Vector Autoregression Model 385
In lag operator notation, the VAR(
 p
) is written as
Π
(
L
)
Y
t
=
c
+
ε
t
where
Π
(
L
) =
I
n
Π
1
L
...
Π
 p
L
 p
. The VAR(
 p
) is stable if the roots of det(
I
n
Π
1
z
· · ·
Π
 p
z
 p
) = 0lie outside the complex unit circle (have modulus greater than one), or,equivalently, if the eigenvalues of the companion matrix
F
=
Π
1
Π
2
· · ·
Π
n
I
n
0
· · ·
00
...
0
...
0 0 I
n
0
have modulus less than one. Assuming that the process has been initializedin the in
nite past, then a stable VAR(
 p
) process is stationary and ergodicwith time invariant means, variances, and autocovariances.If 
Y
t
in (11.1) is covariance stationary, then the unconditional mean isgiven by
µ
= (
I
n
Π
1
· · ·
Π
 p
)
1
c
The
mean-adjusted 
form of the VAR(
 p
) is then
Y
t
µ
=
Π
1
(
Y
t
1
µ
)+
Π
2
(
Y
t
2
µ
)+
· · ·
+
Π
 p
(
Y
t
 p
µ
) +
ε
t
The basic VAR(
 p
) model may be too restrictive to represent su
cientlythe main characteristics of the data. In particular, other deterministic termssuch as a linear time trend or seasonal dummy variables may be requiredto represent the data properly. Additionally, stochastic exogenous variablesmay be required as well. The general form of the VAR(
 p
) model with de-terministic terms and exogenous variables is given by
Y
t
=
Π
1
Y
t
1
+
Π
2
Y
t
2
+
· · ·
+
Π
 p
Y
t
 p
+
Φ
D
t
+
GX
t
+
ε
t
(11.4)where
D
t
represents an (
l
×
1) matrix of deterministic components,
X
t
represents an (
m
×
1) matrix of exogenous variables, and
Φ
and
G
areparameter matrices.
Example 64
Simulating a stationary VAR(1) model using 
S-PLUS 
A stationary VAR model may be easily simulated in
S-PLUS
using the
S+FinMetrics
function
simulate.VAR
. The commands to simulate
=250 observations from a bivariate VAR(1) model
y
1
t
=
0
.
7 + 0
.
7
y
1
t
1
+ 0
.
2
y
2
t
1
+
ε
1
t
y
2
t
= 1
.
3 + 0
.
2
y
1
t
1
+ 0
.
7
y
2
t
1
+
ε
2
t

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->