introduction to var models - nicola · pdf fileintroduction to var models. nicola viegi var...
TRANSCRIPT
1/23Nicola Viegi Var Models
Nicola ViegiUniversity of Pretoria
July 2010
Introduction to VAR ModelsIntroduction to VAR Models
2/23Nicola Viegi Var Models
Introduction
Origins of VAR models
Sims "Macroeconomics and Reality" Econometrica 1980
It should be feasible to estimate large macromodels as unrestricted reduced forms, treating all variables as endogenous
Natural extension of the univariate autoregressive model to multivariate time series
Especially useful for describing the dynamic behaviour of economic and financial time series
Benchmark in forecasting Used for structural inference
3/23Nicola Viegi Var Models
Structural VARStructural VAR
t 10 12 t 11 t 1 12 t 1 yt
t 20 21 t 21 t 1 22 t 1 zt
y b b z y z
z b b y y z− −
− −
= − +γ +γ +ε
= − +γ +γ +ε
• The two variables y and z are endogenous.
• The error terms (structural shocks) εyt and εzt are white noise innovations with standard deviations σy and σz and a zerocovariance.
• Shock εyt affects y directly and z indirectly.• There are 10 (8 coefficients and two standard deviations of the
errors) parameters to estimate.
Consider a bivariate Yt=(yt, zt), first-order VAR model:
4/23Nicola Viegi Var Models
Standard VARStandard VAR
• The structural is not estimable directly• VAR in reduced form is estimable.• In a reduced form representation y and z are just
functions of lagged y and z.• To solve for a reduced form write the structural VAR
in matrix form as:
⎥⎦
⎤⎢⎣
⎡+⎥⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡+⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡
−
−
zt
yt
t
t
t
t
zy
bb
zy
bb
εε
γγγγ
1
1
2221
1211
20
10
21
12
11
ttt YBY ε+Γ+Γ= −110
or, in short
5/23Nicola Viegi Var Models
Standard VARStandard VAR
• Premultipication by B-1 allow us to obtain a standard VAR(1):
• This reduced form can be estimated (by OLS equation by equation)• Before estimating
• Determine the optimal lag length of the VAR• Determine stability conditions (roots of the system inside the unit circle)
• After estimating the reduced form• Hypothesis Testing – Granger Casuality• Impulse Response Function• Variance Decomposition• Identification of Structural VAR
ttt
ttt
ttt
aYAAYBYBBY
YBY
++=+Γ+Γ=
+Γ+Γ=
−
−−
−−
−
110
111
10
1110
ε
ε
6/23Nicola Viegi Var Models
Determining the optimal lag length: Determining the optimal lag length: Information Criterion in a Standard VAR(p)Information Criterion in a Standard VAR(p)
( )
( )
2
2
2AIC ln p (n p n)T
ln(T)SBC ln p (n p n)T
= + +
= + +
Σ
Σ
Information Criteria (IC) can be used to choose the“right” number of lags in a VAR(p): that minimizes IC(p) for p=1, ..., P .
AIC criterion asymptotically overestimates the order with positive probability
SBC criterion estimates the order consistently if the true p is less than the p(max)
7/23Nicola Viegi Var Models
Stability of the system Stability of the system —— roots of the companion matrixroots of the companion matrix
Same concept of stability than in univariate time series
The system is stable if the root of the matrix A1 of the system are all less than 1 in absolute value
ttt aYAAY ++= −110
If not Vector Error Correction Models and Cointegration Models more appropriate.
8/23Nicola Viegi Var Models
GrangerGranger CausalityCausality TestTest
ahead periods , ofForecast t sX
tt
tt
tt
tt
XYYX
sXYsXMSEsXMSE
forecast toeinformativlinearly not is respect to with exogenous is
0 cause-Grangernot does then ))(ˆ())(ˆ( If )2()1(
⇔⇔
>∀=
Granger (1969) “Investigating Causal Relations by Econometric Models and Cross-Spectral Methods”, Econometrica, 37
Consider two random variables tt YX ,
,....),,....,|()(ˆ,....),|()(ˆ11
)2(1
)1(−−+−+ == ttttsttttstt YYXXXEsXXXXEsX
2))(ˆ())(ˆ( sXXEsXMSE tstt −= +Define Minimum square error
9/23Nicola Viegi Var Models
Assume a lag length of p
tptpttptpttt aXYYXXXcX +++++++= −−−−−− βββααα ......... 221122111
Estimate by OLS and test for the following hypothesis
0any :) cause-Grangernot does ( 0......:
1
210
≠
====
i
ttp
HXYH
β
βββ
Unrestricted sum of squared residuals
Restricted sum of squared residuals
∑=t
taRSS 21 ˆ
∑=t
taRSS 22
ˆ
( )12,,
1
12
ifreject )12/(
/)(
−−>−−
−=
pTpFFpTRSS
pRSSRSSF
α
Important – Stationarity of the data
GrangerGranger CausalityCausality TestTest
10/23Nicola Viegi Var Models
Impulse Response FunctionImpulse Response Function
Objective: Show the reaction of the system to a shock
........: at time Redating
)]([)(
....)(,stationary-covariance is system theIf
....
112211
12211
2211
+Ψ+Ψ++Ψ+Ψ++=+
Φ=Ψ
+Ψ+Ψ++=Ψ+=
+Φ++Φ+Φ+=
−+−+−+++
−
−−
−−−
tstsstststst
ttttt
tptpttt
aaaaaYst
LL
aaaaLY
aYYYcY
µ
µµ
[ ])(,
)(
'
sij
jt
sti
sijs
t
st
ayaY
ψ
ψ
=∂∂
=Ψ=∂∂
+
+
n x n
Reaction of the i-variable to a unit changein innovation j
(multipliers)
11/23Nicola Viegi Var Models
Impulse Response FunctionImpulse Response Function
Impulse-response function: response of to one-time impulse in with all other variables dated t or earlier held constant.
stiy +,
jty
ijjt
sti
ay
ψ=∂∂ +,
s
ijψ
1 2 3
12/23Nicola Viegi Var Models
ExampleExample: IRF : IRF forfor a VAR(1)a VAR(1)
⎥⎥⎦
⎤
⎢⎢⎣
⎡=Σ⎥
⎦
⎤⎢⎣
⎡+⎥
⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡
−
−
2212
122
1
2
1
2
11
2221
1211
2
1 ;1 σσ
σσφφφφ
at
tt
t
t
aa
yy
yy
t
occur) shocks more no(unit) 1by increases (10
00
220
21
t
tt
yatyyt
====<
Reaction of the system
⎥⎦
⎤⎢⎣
⎡Φ=⎥
⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡
⎥⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡
⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡
⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡
10
10
10
10
10
12221
1211
2
1
2
2221
1211
21
11
2221
1211
22
12
22
11
2221
1211
21
11
20
10
ss
s
s
yy
yy
yy
yy
yy
φφφφ
φφφφ
φφφφ
φφ
φφφφ
M
(impulse)
13/23Nicola Viegi Var Models
ForecastForecast Error Error VarianceVariance DecompositionDecomposition
Contribution of the j-th orthogonalized innovation to the MSE of the s-period ahead forecast
If shocks does not explain none of the forecast error variance of at all forecast horizon we can say that the sequence is exogenous
14/23Nicola Viegi Var Models
Identification in a Standard VAR(1)Identification in a Standard VAR(1)
ytt 10 t 111 1212
t 20 21 22 t 1 zt
y b y1 bz b z0 1
−
−
εγ γ ⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎡ ⎤⎡ ⎤= + + ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥⎢ ⎥ γ γ ε⎣ ⎦ ⎣ ⎦⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦
Is it possible to recover the parameters in the structural VAR from the estimated parameters in the standard VAR? No!!
There are 10 parameters in the bivariate structural VAR(1) and only 9 estimated parameters in the standard VAR(1).
The VAR is underidentified.
If one parameter in the structural VAR is restricted the standard VAR is exactly identified.
Sims (1980) suggests a recursive system to identify the model letting b21=0. Choleski decomposition.
15/23Nicola Viegi Var Models
Identification in a Standard VAR(1) Identification in a Standard VAR(1)
ytt 10 t 111 1212 12 12
t 20 21 22 t 1 zt
t 10 t 1 1t11 12
t 20 21 22 t 1 2t
y b y1 b 1 b 1 bz b z0 1 0 1 0 1
y a y ea az a a a z e
−
−
−
−
εγ γ− − − ⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤= + + ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥γ γ ε⎣ ⎦ ⎣ ⎦ ⎣ ⎦⎣ ⎦⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎡ ⎤= + +⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣ ⎦⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦
The parameters of the structural VAR can now be identified from the following 9 equations
2 2 210 10 12 20 20 20 1 y 12 z
211 11 12 21 21 21 2 z
212 12 12 22 22 22 1 2 12 z
a b b b a b var(e ) b
a b a var(e )a b a cov(e ,e ) b
= − = =σ + σ
= γ − γ = γ =σ
= γ − γ = γ = − σ
b21=0 implies
16/23Nicola Viegi Var Models
Identification in a Standard VAR(1)Identification in a Standard VAR(1)
Both structural shocks can now be identified
b21=0 implies y does not have a contemporaneous effect on z.
Both εyt and εzt affect y contemporaneously but only εzt affects z contemporaneously.
The residuals of e2t are due to pure shocks to z.
There are other methods used to identify models – Restrictions coming from theory – (Sims Bernake, Blanchard and Quah etc)
17/23Nicola Viegi Var Models
CriticsCritics onon VARVAR
A VAR model can be a good forecasting model, but it is an atheoreticalmodel (as all the reduced form models are).
To calculate the IRF, the order matters: Identification not unique.
Sensitive to the lag selection
Dimensionality problem.
Standard Tool for Macroeconomic Analysis
18/23Nicola Viegi Var Models
An example of VAR analysisAn example of VAR analysis
Leeper Sims and Zha (1996) “What does monetary policy do”
19/23Nicola Viegi Var Models
The model
..
Money M Income, X Prices,
2
2
2
2
1
1
1
1
t
t
pt
pt
pt
p
t
t
t
t
t
t
t
t
t
t
t
tt
a
M
X
P
MXP
MXP
cYMXP
Y
P
+⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
Φ+⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡Φ+
⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡Φ+=
⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡=
===
−
−
−
−
−
−
−
−
−
)1(
333231
232221
131211
1
⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡=Φ
φφφφφφφφφ
⎩⎨⎧
≠=Ω
==ττ
τ tt
aaEaE tt 0)'(0)(
Where And
Steps: • Optimal Lag Length• Stability• Residual Analysis• Identification (if interest in structural form or in structural shocks)• Impulse Response Function • Variance Decomposition
21/23Nicola Viegi Var Models
-1-.
50
.51
Imag
inar
y
-1 -.5 0 .5 1Real
Roots of the companion matr ix
Command varstable, graph after var or svar
Stability of the VAR — Roots of the companion matrix
22/23Nicola Viegi Var Models
Granger Causality Test
Command – vargranger after varbasic, var or svar
23/23Nicola Viegi Var Models
-.005
0
.005
.01
.015
-.005
0
.005
.01
.015
-.005
0
.005
.01
.015
0 50 0 50 0 50
varbasic, lm2, lm2 varbasic, lm2, lp varbasic, lm2, ly
varbasic, lp, lm2 varbasic, lp, lp varbasic, lp, ly
varbasic, ly, lm2 varbasic, ly, lp varbasic, ly, ly
95% CI orthogonalized irf
step
Graphs by irfname, impulse variable, and response variable
Cholesky Decomposition – Order Prices/Income/Money
varbasic lp ly lm2, lags(1/6) step(50) oirf
24/23Nicola Viegi Var Models
0
.5
1
0
.5
1
0
.5
1
0 50 0 50 0 50
varbasic, lm2, lm2 varbasic, lm2, lp varbasic, lm2, ly
varbasic, lp, lm2 varbasic, lp, lp varbasic, lp, ly
varbasic, ly , lm2 varbasic, ly, lp varbasic, ly, ly
95% CI fraction of mse due to impulse
step
Graphs by irfname, impulse variable, and response variable
Forecast Error Variance Decomposition - FEVD
varbasic lp ly lm2, lags(1/6) step(50) fevd