You are on page 1of 5

System of Equations: EC309

William Matcham August 2, 2013

System of Equations
(a) The Kronecker product of matrices A and B is the pr qs matrix:
pq r s

1. Kronecker product and vec operator

a11 . . . a1q . .. . , B = A= . . . . . ap1 . . . apq

b11 . . . b1s . . .. . . . . . br1 . . . brs

a11 b11 . . . a11 b1s . . . . ... . . a11 br1 . . . a11 brs . a11 B . . . a1q B . . . . . . . . . . . . . AB = . = . . . . . . . . . . . . ap1 B . . . apq B a b p1 11 . . . ap1 b1s . . . . . ... . ap1 br1 . . . ap1 brs RULES (b) Associative Law: (A B ) C = A (B C ) = A B C (c) Distributive Law: (A + B ) C = (A C ) + (B C )

...... ...... ...... ...... ...... ...... ...... ......

a1q b11 . . . a1q b1s . . . . . ... . a1q br1 . . . a1q brs . . . . . . . . . . . . . . . . . . apq b11 . . . apq b1s . . . . . ... . apq br1 . . . apq brs

(d) (A B )(C D) = AC BD, provided that AC and BD are dened. (e) (A B ) = A B (f) (A B )1 = A1 B 1 , provided that det(A) = 0 and det(B ) = 0 These can all be shown from rst principles.

(g) Note the dierence of (d),(e) and (f) from the standard matrix product. (h) For a matrix A with j th column denoted aj : . . . A = a1 . . . . . . We dene the vec operator by . . . aj . . . . . . . . . aq . . .

. . . a1 . . . . . . vec(A) = aj . . . . . . aq . . .

This turns an N q matrix into an N q 1 vector by stacking the columns of the matrix vertically. (i) vec(ABC ) = (C A)vec(B ).

2.1

Notational Convention

1. Say you have N equations i = 1, ..., N . For example, suppose we are estimating the demand functions for a selection of N dierent goods. 2. Observation in each equation taken at time t = 1, ..., T . y i1 y1t . . . . . . 3. We can either write yt = yit or yi = yit . . . . . . yN t yiT
N 1 T 1

4. yt gives all y values for equations 1 through to N at time t, whereas yi gives all the values of y in equation i for times t = 1 through to t = T .

1t i 1 . . . . . . 5. Likewise we either write t = it or i = it . . . . . . N t iT


N 1 T 1

6. Say we have k regressors, so that xt = (x1t , . . . , xkt ) and i = (i1 , . . . , ik ) are k dimensional vectors. 7. i is a k 1 vector, giving all slopes ij for j = 1, . . . , k for equation i

2. Multivariate Regression

A mulitvariate regression is a collection of N regressions, with the SAME set of regressors in each equation. So: yit = xt i + it i = 1, . . . N ; t = 1, . . . T

We now set up a matrix of y, , and x values, since we have these values either across time, across regressors, or across equations. We let the Matrix B stack the row vectors i : ... 1 . . . i . . . N ... ... ... ... ... 11 . . . 1k . . . . . . . . . = i1 . . . ik . . . . . . . . . N 1 . . . N k
N k

... B = ... ... ... Hence B is a N k matrix. ...

... X = ... ... ... Hence X is a T k matrix.

x1 . . . . . . ... xt . . . . . . ... xT

x11 . . . x1k . . . . . . . . . = xt1 . . . xtk . . . . . . . . . ... xT 1 . . . x T k


T k

...

... Y = ... ... ... Hence Y is a T N matrix.

y1 . . . . . . ... yt . . . . . . ... yT

y11 . . . yi1 . . . . . . . . . . . . . . . y . . . y . . . = 1t it . . . . . . . . . . . . ... y1T . . . yiT . . .


T N

yN 1 . . . yN t . . . yN T

...

... = ... ... ... Hence is a T N matrix.

1 . . . . . . ... t . . . = . . . ... T . . .

11 . . . . . . . . . 1t . . . . . . . . . 1T . . .

i1 . . . . . . . . . it . . . . . . . . . iT T N ...

N 1 . . . N t . . . N T

Then the multivariate regression model is written either: yt N 1 Or yi T 1 = X i + i


T k k1

= B xt + t
N k k1

t = 1, . . . , T

N 1

i = 1, . . . , N

T 1

For each equation i, we have all values of equation i at times t = 1, . . . , T We wish to write the whole system in one formula, and this is given by:
T N

Y = X B +
T k kN

T N

Every column of Y,

and XB correspond to one equation.

Another way to write this is: . . . y1 . . . . . . y2 . . . . . . . . . y N . . .

X 0 ... 0 0 X ... 0 . . .. . . . . . . . . 0 0 ... X


N T N k

N . . .
N k1

. . . 1 . . . . . . 2 . . . . . . . . .

+ N . . .

. . . 1 . . . . . . 2 . . . . . . . . .

N T 1

N T 1

The two above representations can be connected by X 0 ... y1 0 X ... . . . . . = .. . . . . . yN 0 0 ... y


N T 1 N N

the Kronecker product as follows: 0 1 1 0 . . . . . + . . . . N N X


N T 1

= ( IN X ) +
T k N k1

Where y = vec(Y ), = vec( ) and = vec(B ) Another way realise this is to consider: Y = XB IN + Now vec both sides, to give: vec(Y ) = vec(XB IN + ) y = vec(XB IN ) + vec( ) = (IN X )vec(B ) + = (IN X ) + Using RULE (i) on the penultimate line.

You might also like