Scribd
Upload
231 views4 pages

The Impulse-Response Function Lecture Notes

The document discusses the orthogonalized impulse-response function for a stationary VAR(p) model. It provides three key points: 1) A VAR model cannot be used correctly for dynamic simulations unless the researcher interprets the instantaneous correlations between error terms. This is called "structuralizing" the VAR. 2) Sims' original proposal orthogonalizes the VAR using a triangular factorization of the error covariance matrix. This yields uncorrelated error terms whose variances can be interpreted as shocks to individual equations. 3) The orthogonalized impulse response function shows the response of variable y to a one-time shock in another variable u, holding all past variables constant.

Uploaded by

runawayyy
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
231 views4 pages

The Impulse-Response Function Lecture Notes

The document discusses the orthogonalized impulse-response function for a stationary VAR(p) model. It provides three key points: 1) A VAR model cannot be used correctly for dynamic simulations unless the researcher interprets the instantaneous correlations between error terms. This is called "structuralizing" the VAR. 2) Sims' original proposal orthogonalizes the VAR using a triangular factorization of the error covariance matrix. This yields uncorrelated error terms whose variances can be interpreted as shocks to individual equations. 3) The orthogonalized impulse response function shows the response of variable y to a one-time shock in another variable u, holding all past variables constant.

Uploaded by

runawayyy
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

The Impulse-Response function Lecture Notes

Eduardo Rossi May 26, 2004


THE ORTHOGONALIZED IMPULSE-RESPONSE FUNCTION Stationary V AR(p): (L) yt = t t V W N (0, ) A stationary VAR model has to be considered as a reduced form model where no explanations of the instantaneous relationship among variables are provided. V MA () representation: yt = Impulse-response function yt+n =
X i=0 X i=0

i ti

0 = In

i t+ni

{n }i,j =

yit+n jt

the response of yt+n to a one-time impulse in yjt with all other variables dated t or earlier held constant. 1

Remember that is not diagonal, i.e. jt is correlated with mt . There are indirect eects through mtn on yit which is yit+n mt It is not legitimate to perturb one element of t , leaving the others to zero. Therefore, a VAR cannot correctly used for dynamic simulations unless the reasercher is ready to provide an interpretation of the instantaneous correlations among the elements of t . This interpretation is called structuralisation of the VAR. Sims (1980) original proposal consisted in moving from a non-orthogonal to an orthogonalised VMA representation via triangular factorization of the matrix : = ADA0 where 0 0 ... . . . 1

Using the matrix A

np p o D = diag d11 , , dnn ut = A1 t

A=

1 a21 . . .

0 1 . . .

an1 an2

with E (ut ) = A1 E (t ) = 0 E (ut u0t ) = A1 E (t 0t ) A1 0 1 1 = A A 0 1 1 0 = A ADA A = D 2


0

The elements in ut are mutually uncorrelated with variance: djj = E u2 var (ut ) t The system appears as: u1t = 1t u2t = a21 1t 2t . . . unt = an1 1t an2 2t . . . nt uit cannot be interpreted as the shock to the i-th variable given the past of the system, but as the shock to the economic relation which is represented by i-th equation, in fact A1 (L) yt = A1 t (L) yt = ut (L) = + L + L2 + . . . + Lp 0 1 2 p where = A1 0 is lower triangular with unit diagonal elements, such that uit = yit E yit xt1 , y1t , . . . , y(i1)t The V M A () representation is 0 0 x0t1 = yt1 , . . . , ytp yt =
X i=0

i Auti

The orthogonalized impulse-response function is yit+n = {i A}ij ujt 3

For a given observed sample of size T , we estimate 1 , . . . , p by OLS Then we estimate b = (1/T )
T X t=1

b bb b Finally, = ADA0 .

btb0t

Alternatively, we can use the Choleski factorization of 0 = P P 0 = AD1/2 AD1/2 P 1 P 1 = In


0

Premultiply the system by the inverse of the Cholesky factor P 1 : P 1 (L) yt = P 1 t (L) yt = et E (et ) = 0 0 E (et e0t ) = P 1 E (t 0t ) P 1 = In

You might also like