Vector Autoregressions I
Empirical Macroeconomics - Lect 1
January 2012
properties:
E(t ) = 0 (mean zero);
E(t t ) = (full variance-covariance matrix);
c1
a1,11 a1,12
y1t
y2t = c2 + a1,21 a1,22
y3t
c3
a1,31 a1,32
y1 = c + A1 y0 + 1
y1t1
a1,13
a1,23 y2t1 +
a1,33
y3t1
y1t2
1t
y2t2 + 2t
y3t2
3t
y2 = c + A1 y1 + 2 = c + A1 (c + A1 y0 + 1 ) + 2
yt
= (Im + A1 )c + A21 y0 + A1 1 + 2
..
.
1 i
= (Im + A1 + ... + A1t1 )c + At1 y0 + it=
0 A1 ti
t ,
yt = + i=0 Ai1 ti
where = E(yt ) is the unconditional mean computed as
(Im A1 )1 c.
yt can be decomposed between a deterministic term ()
or:
A(L) = Im A1 L A2 L2 ... Ap Lp
A ( L ) yt = c + t .
Now dene a operator (L) = i=0 i Li such that:
( L ) A ( L ) = Im
that is, (L) = A(L)1 . This operator only works if the
inverse of A(L) exists that requires that
det(Im A1 z ... Ap zp ) = 0, which also implies that the
VAR is stable.
yt = ( L ) c + ( L ) t
= (i=0 i ) c + i=0 i ti
which is the MA() representation of a VAR(p).
Algebra can be used to show that how to compute (L)
out of A(L):
0 = Im
i =
ij=1
yt = c +
.5 .1
.4 .5
= (Im A1 ... Ap )1 c.
And that the MA representation can be simplied to:
yt = + t + 1 t1 + 2 t2 + ...
yt1 +
1 = A1 =
.5 .1
.4 .5
mean and variance do not depend on time, that is, they are
constant over time.
A stable VAR(p) process is stationary.
yt 2 + t .
A21
3 = 2 A1 + 1 A2 =
+ A2 =
A31
.29 .1
.65 .29
+ A2 A1 + A1 A2 =
.21 .079
.566 .21
..
.
i = i1 A1 + i2 A2
Wold Decomposition I
Macroeconomists like to use VARs to represent the empirical
joint empirical process of macroeconomic time series. An
important support for this is the Wold decomposition.
0 0
.25 0
2 = 1 A1 + A2 =
ij Aj
Wold Decomposition II
vt P1 t
of =
PP .
1
0 0
A = a21 1 0
a31 a32 1
0
u1 0
D1/2 = 0 u2 0
0
0 u3
D = diag(2u1 , 2u2 , 2u3 )
0
0
u1
u2
0
P = a21 u1
a31 u1 a32 u2 u3
21
cov(1 , 2 ) cov(1 , 3 )
22
cov(2 , 3 ) .
= cov(1 , 2 )
cov(1 , 3 ) cov(2 , 3 )
23
Recursive VAR I
To understand better the difference between the structural
Aut = t
u1t
1
0 0
1t
a21 1 0 u2t = 2t .
u3t
3t
a31 a32 1
The implied equations (that allow the computation of are
ut s) are:
u1t = 1t
u2t = 2t a21 u1t
u3t = 3t a31 u1t a32 u2t .
Recursive VAR II
t = 0; a21
t = 1; 1,21 a21
t = 2; 2,21 a21
..
.
Forecast errors I
Forecast errors II
Variance Decomposition I
Variance Decomposition II
Using this representation, we can compute the MSE(y t+s )
using ujt :
MSE(y t+s ) = m
j=1 {var(ujt )[aj aj + 1 aj aj 1 +
+... + s1 aj aj s 1 ].
Econometric software normally computes the proportion