2.

5 Variance decomposition and innovation

accounting

Consider the VAR(p) model

(L)yt = t,
where

(L) = Im 1L 2L2 pLp

is the lag polynomial of order p with m m
coecient matrices i, i = 1, . . . p.

Provided that the stationarity condition holds

we may obtain a vector MA representation of
yt by left multiplication with 1(L) as

yt = 1(L) t = (L) t
where

1(L) = (L) = Im + 1L + 2L2 + .

78
The m m coecient matrices 1, 2, . . .
may be obtained from the identity
p
(L)(L) = (Im iLi)(Im+ iLi) = Im
i=1 i=1
as
j
j = jii
i=1
with 0 = Im and j = 0 when i > p, by
multiplying out and setting the resulting co-
ecient matrix for each power of L equal to
zero. For example, start with L11 = L1:

1L1 + 1L1 = (1 1)L1 0

1
1 = 1 = 01 = 1ii
i=1
Consider next L2
1:

2L2 2 2
1 11L1 2L1 0
2
2 = 11 + 2 = 2ii
i=1
j
The result generalizes to any power L1, which
yields the transformation formula given above.
79
Now, since

yt+s = (L) t+s = t+s + i t+si
i=1
we have that the eect of a unit change in
t on yt+s is

yt+s
= s.
t
Now the t's represent shocks in the sys-
tem. Therefore the i matrices represent
the model's response to a unit shock (or in-
novation) at time point t in each of the vari-
ables i periods ahead. Economists call such
parameters dynamic multipliers.

The response of yi to a unit shock in yj is

therefore given by the sequence below, known
as the impulse response function,

ij,1, ij,2, ij,3, . . . ,

where ij,k is the ijth element of the matrix
k (i, j = 1, . . . , m).

80
For example if we were told that the rst el-
ement in t changes by 1 at the same time
that the second element changed by 2, . . . ,
and the mth element by m, then the com-
bined eect of these changes on the value of
the vector yt+s would be given by
yt+s yt+s
yt+s = 1 + + m = s,
1t mt
where = (1, . . . , m).

Generally an impulse response function traces

the eect of a one-time shock to one of the
innovations on current and future values of
the endogenous variables.

81
Example: Exogeneity in MA representation

Suppose we have a bivariate VAR system

such that xt does not Granger cause yt. Then
we can write

(1)
yt 0 yt1
= 11 (1) (1) +
xt 22 xt1
21
(p)
11 0
+ (p) ytp + 1,t .
(p) xtp
21 22 2,t
j
Then the coecient matrices j = i=1 jii
in the corresponding MA representation are
lower triangular as well (exercise):

(i)
yt 1,t 11 0 1,ti
= + (i) (i)
xt 2,t 2,ti
i=1 21 22
Hence, we see that variable y does not react
to a shock in x.

82
Ambiguity of impulse response functions

Consider a bivariate VAR model in vector MA

representation, that is,

yt = (L) t with E( t t) = ,
where (L) gives the response of yt = (yt1, yt2)
to both elements of t, that is, t1 and t2.

Just as well we might be interested in evalu-

ating responses of yt to linear combinations
of t1 and t2, for example to unit move-
ments in t1 and t2 + 0.5 t1. This may be
done by dening new shocks t1 = t1 and
t2 = t2 + 0.5 t1, or in matrix notation

1 0
t = Q t with Q= .
0.5 1

83
The vector MA representation of our VAR in
terms of the new chocks becomes then

yt = (L) t = (L)Q1Q t =: (L)t

with
(L) = (L)Q1.
Note that both representations are observa-
tionally equivalent (they produce the same
yt), but yield dierent impulse response func-
tions. In particular,

0 = 0 Q1 = I Q1 = Q1,
which implies that single component shocks
may now have contemporaneous eects on
more than one component of yt. Also the co-
variance matrix of residuals will change, since

E(tt) = E(Q t tQ ) = unless Q = I.

But the fact that both representations are
observationally equivalent implies that it is
our own choice which linear combination of
the ti's we nd most useful to look at in the
response analysis!
84
Orthogonalized impulse response functions

Usually the components of t are contem-

poraneously correlated. For example in our
VAR(2) model of the equity-bond data the
contemporaneous residual correlations are
=================================
FTA DIV R20 TBILL
---------------------------------
FTA 1
DIV 0.123 1
R20 -0.247 -0.013 1
TBILL -0.133 0.081 0.456 1
=================================

If the correlations are high, it doesn't make

much sense to ask "what if t1 has a unit im-
pulse" with no change in t2 since both come
usually at the same time. For impulse re-
sponse analysis, it is therefore desirable to ex-
press the VAR in such a way that the shocks
become orthogonal, (that is, the ti's are un-
correlated). Additionally it is convenient to
rescale the shocks so that they have a unit
variance.
85
So we want to pick a Q such that E(tt) = I.
This may be accomplished by coosing Q such
that Q1Q1 = , since then

E(tt) = E(Q t tQ ) = E(Q Q ) = I.

Unfortunately there are many dierent Q's,
whose inverse S = Q1 act as "square roots"
for , that is, SS = .

This may be seen as follows. Choose any

orthogonal matrix R (that is, RR = I) and
set S = SR. We have then

SS = SRR S = SS = .

Which of the many possible S's, respectively

Q's, should we choose?

86
Before turning to a clever choice of Q (resp. S),
let us briey restate our results obtained so
far in terms of S = Q1.

If we nd a matrix S such that SS = , and

transform our VAR residuals such that

t = S1 t,
then we obtain an observationally equivalent
VAR where the shocks are orthogonal (i.e.
uncorrelated with a unit variance), that is,
1 1
E(tt) = S1E( t t)S = S1 S = I.
The new vector MA representation becomes

yt = (L)t = iti,
i=0
where i = iS (m m matrices) so that
0 = S = Im. The impulse response function
of yi to a unit shock in yj is then given by
the othogonalised impulse response function
, , , . . . .
ij,0 ij,1 ij,2

87
Choleski Decomposition and Ordering of Variables

Note that every orthogonalization of corre-

lated shocks in the original VAR leads to con-
temporaneous eects of single component
shocks ti to more than one component of
yt, since 0 = S will not be diagonal unless
was diagonal already.

One generally used method in choosing S

is to use Cholesky decomposition which re-
sults in a lower triangular matrix with positive
main diagonal elements for 0, e.g.

(0)
y1,t 0 1,t (1)
= 11
(0) (0) + t1+. . .
y2,t 21 22 2,t

Hence Cholesky decomposition of implies

that the second shock 2,t does not aect
the rst variable y1,t contemporaneously, but
both shocks can have a contemporaneous ef-
fect on y2,t (and all following variables, if we
had choosen an example with more than two
components). Hence the ordering of vari-
ables is important!
88
It is recommended to try various orderings
to see whether the resulting interpretations
are consistent. The principle is that the rst
variable should be selected such that it is the
only one with potential immediate impact on
all other variables. The second variable may
have an immediate impact on the last m 2
components of yt, but not on y1t, the rst
component, and so on. Of course this is
usually a dicult task in practice.

Variance decomposition

The uncorrelatedness of the t's allows to

decompose the error variance of the s step-
ahead forecast of yit into components ac-
counted for by these shocks, or innovations
(this is why this technique is usually called
innovation accounting). Because the inno-
vations have unit variances (besides the un-
correlatedness), the components of this error
variance accounted for by innovations to yj
s1 (l)2
is given by l=0 ij , as we shall see below.
89
Consider an orthogonalized VAR with m com-
ponents in vector MA representation,

yt = (l)tl .
l=0
The s step-ahead forecast for yt is then

Et(yt+s) = (l)t+sl .
l=s
Dening the s step-ahead forecast error as

et+s = yt+s Et(yt+s)

we get
s1
et+s = (l)t+sl .
l=0
It's i'th component is given by
s1 m m s1
(l) (l)
ei,t+s = ij j,t+sl = ij j,t+sl .
l=0 j=1 j=1 l=0

90
Now, because the shocks are both serially
and contemporaneously uncorrelated, we get
for the error variance
m s1
(l)
V(ei,t+s) = V(ij j,t+sl )
j=1 l=0
m s1
(l)2
= ij V(j,t+sl ).
j=1 l=0
Now, recalling that all shock components have
unit variance, this implies that

m s1
(l)2
V(ei,t+s) = ij ,
j=1 l=0
s1 (l)2
where l=0 ij accounts for the error vari-
ance generatd by innovations to yj , as claimed.

Comparing this to the sum of innovation re-

sponses we get a relative measure how im-
portant variable js innovations are in the ex-
plaining the variation in variable i at dierent
step-ahead forecasts, i.e.,
s1 (l) 2
2 = 100 l=0 ij
Rij,l 2
.
m s1 (l)
k=1 l=0 ik
91
Thus, while impulse response functions traces
the eects of a shock to one endogenous
variable on to the other variables in the VAR,
variance decomposition separates the varia-
tion in an endogenous variable into the com-
ponent shocks to the VAR.

Example. Let us choose in our example two

orderings. One according to the feedback
analysis

[(I: FTA, DIV, R20, TBILL)],

and an ordering based upon the relative tim-
ing of the trading hours of the markets

[(II: TBILL, R20, DIV, FTA)].

In EViews the order is simply dened in the
Cholesky ordering option. Below are results
in graphs with I: FTA, DIV, R20, TBILL;
II: R20, TBILL DIV, FTA, and III: General
impulse response function.

92
Impulse responses:

Order {FTA, DIV, R20, TBILL}

Response to Cholesky One S.D. Innovations 2 S.E.
Response of DFTA to DFTA Response of DFTA to DDIV Response of DFTA to DR20 Response of DFTA to DTBILL
7 7 7 7
6 6 6 6

5 5 5 5
4 4 4 4
3 3 3 3
2 2 2 2

1 1 1 1
0 0 0 0

-1 -1 -1 -1
-2 -2 -2 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DDIV to DFTA Response of DDIV to DDIV Response of DDIV to DR20 Response of DDIV to DTBILL
1.6 1.6 1.6 1.6

1.2 1.2 1.2 1.2

0.8 0.8 0.8 0.8

0.4 0.4 0.4 0.4

0.0 0.0 0.0 0.0

-0.4 -0.4 -0.4 -0.4

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DR20 to DFTA Response of DR20 to DDIV Response of DR20 to DR20 Response of DR20 to DTBILL
4 4 4 4

3 3 3 3

2 2 2 2

1 1 1 1

0 0 0 0

-1 -1 -1 -1

-2 -2 -2 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DTBILL to DFTA Response of DTBILL to DDIV Response of DTBILL to DR20 Response of DTBILL to DTBILL
6 6 6 6

5 5 5 5

4 4 4 4

3 3 3 3

2 2 2 2

1 1 1 1

0 0 0 0

-1 -1 -1 -1

-2 -2 -2 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Order {TBILL, R20, DIV, FTA}

Response of DFTA to DFTA
7 7
6 6
5 5
4 4
3 3
2 2
1 1
Response to Cholesky One S.D. Innovations 2 S.E.
Response of DFTA to DDIV Response of DFTA to DR20 Response of DFTA to DTBILL
7 7
6 6
5 5
4 4
3 3
2 2
1 1

0 0 0 0
-1 -1 -1 -1
-2 -2 -2 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DDIV to DFTA Response of DDIV to DDIV Response of DDIV to DR20 Response of DDIV to DTBILL
1.6 1.6 1.6 1.6

1.2 1.2 1.2 1.2

0.8 0.8 0.8 0.8

0.4 0.4 0.4 0.4

0.0 0.0 0.0 0.0

-0.4 -0.4 -0.4 -0.4

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DR20 to DFTA Response of DR20 to DDIV Response of DR20 to DR20 Response of DR20 to DTBILL
4 4 4 4

3 3 3 3

2 2 2 2

1 1 1 1

0 0 0 0

-1 -1 -1 -1

-2 -2 -2 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DTBILL to DFTA
7 7
6 6
5 5
4 4
3 3
Response of DTBILL to DDIV Response of DTBILL to DR20 Response of DTBILL to DTBILL
7 7
6 6
5 5
4 4
3 3

2 2 2 2
1 1 1 1

0 0 0 0
-1 -1 -1 -1
-2 -2 -2 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

93
Impulse responses continue:

Response to Generalized One S.D. Innovations 2 S.E.

Response of DFTA to DFTA Response of DFTA to DDIV Response of DFTA to DR20 Response of DFTA to DTBILL
8 8 8 8

6 6 6 6

4 4 4 4

2 2 2 2

0 0 0 0

-2 -2 -2 -2

-4 -4 -4 -4
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DDIV to DFTA Response of DDIV to DDIV Response of DDIV to DR20 Response of DDIV to DTBILL
1.6 1.6 1.6 1.6

1.2 1.2 1.2 1.2

0.8 0.8 0.8 0.8

0.4 0.4 0.4 0.4

0.0 0.0 0.0 0.0

-0.4 -0.4 -0.4 -0.4

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DR20 to DFTA Response of DR20 to DDIV Response of DR20 to DR20 Response of DR20 to DTBILL
4 4 4 4

3 3 3 3

2 2 2 2

1 1 1 1

0 0 0 0

-1 -1 -1 -1

-2 -2 -2 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of DTBILL to DFTA Response of DTBILL to DDIV Response of DTBILL to DR20 Response of DTBILL to DTBILL
7 7 7 7
6 6 6 6

5 5 5 5

4 4 4 4

3 3 3 3
2 2 2 2

1 1 1 1
0 0 0 0

-1 -1 -1 -1

-2 -2 -2 -2
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

The general impulse response function are

dened as
GI(j, i, Ft1) = E[yt+j | it = i, Ft1]E[yt+j |Ft1].
That is dierence of conditional expectation
given an one time shock occurs in series j.
These coincide with the orthogonalized im-
pulse responses if the residual covariance ma-
trix is diagonal.
Pesaran,
M. Hashem and Yongcheol Shin (1998).
Impulse Response Analysis in Linear Multivariate
Models, Economics Letters, 58, 17-29.
94
Variance decomposition graphs of the equity-
bond data

Variance Decomposition 2 S.E.

Percent DFTA variance due to DFTA Percent DFTA variance due to DDIV Percent DFTA variance due to DR20 Percent DFTA variance due to DTBILL
120 120 120 120

100 100 100 100

80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0

-20 -20 -20 -20

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Percent DDIV variance due to DFTA Percent DDIV variance due to DDIV Percent DDIV variance due to DR20 Percent DDIV variance due to DTBILL
120 120 120 120

100 100 100 100

80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0

-20 -20 -20 -20

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Percent DR20 variance due to DFTA Percent DR20 variance due to DDIV Percent DR20 variance due to DR20 Percent DR20 variance due to DTBILL
100 100 100 100

80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0

-20 -20 -20 -20

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Percent DTBILL variance due to DFTA Percent DTBILL variance due to DDIV Percent DTBILL variance due to DR20 Percent DTBILL variance due to DTBILL
100 100 100 100

80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0

-20 -20 -20 -20

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

95
On estimation of the impulse response coecients

Consider the VAR(p) model

(L)yt = t,
with (L) = Im 1L 2L2 pLp.
Then under stationarity the vector MA rep-
resentation is

y = + 1 t1 + 2 t2 +
When we have estimates of the AR-matrices
i denoted by ^ i, i = 1, . . . , p; the next prob-
lem is to construct estimates ^ j for the MA
matrices j . Recall that
j
j = jii
i=1
with 0 = Im, and j = 0 when i > p. The
estimates ^ j can be obtained by replacing
^ i.
the i's by their corresponding estimates

96
Next we have to obtain the orthogonalized
impulse response coecients. This can be
done easily, for letting S be the Cholesky de-
composition of such that

= SS ,
we can write
yt =
i=0 i ti
= SS1
i=0 i ti
=
i=0 i ti,
where
i = iS
and t = S1 t. Then
1
Cov(t) = S1 S = I.
The estimates for i are obtained by re-
^ t and using
placing t with their estimates
Cholesky decomposition of ^ .

97