MIT OpenCourseWare

http://ocw.mit.edu

14.384 Time Series Analysis

Fall 2008

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
Notation and some Linear Algebra 1

14.384 Time Series Analysis, Fall 2007

Professor Anna Mikusheva
Paul Schrimpf, scribe
September 27, 2007

Lecture 7

VARs

Notation and some Linear Algebra

Let
p
X
yt = aj yt−j + et (1)
j=1

where yt and et are k × 1, and aj is k × k. et is white noise with Eet e0t = Ω and Eet e0s = 0
³ Pp ´ ³ Pp ´
Lemma 1. yt is stationary if det Ik − j=1 aj z j 6= 0 for all |z| ≤ 1, i.e. all roots of det Ik − j=1 aj z j
are outside the unit circle.
Definition 2. Companion form of (1) is:
 
  a1 a2 ... ap  
yt et
 I 0 ... 0 
 yt−1     0 
Yt =   ,A= 0 I 0 ...  , Et = 
... ...
     
.. ..

   ..   
 . . . 
yt−p+1 0
I 0

so Yt and Et are kp × 1 and A is kp × kp. We can rewrite (1) as

Yt =AYt−1 + Et
⇒
ΣY =AΣY A0 + ΣE
 
a11 a12
Definition 3. If A =  a21 a22 , then
a31 a32
 
a11

 a12 

 a21 
vec(A) =  

 a22 

 a31 
a32

A useful property is
Lemma 4.

vec(ABC) = (C 0 ⊗ A)vec(B)
Estimation 2

so,
vec(ΣY ) =vec (AΣY A0 + ΣE )
=(A ⊗ A)vec(ΣY ) + vec(ΣE )
⇒
vec(ΣY ) = (I − (A ⊗ A))−1 vec(ΣE )

Estimation
ˆ=
Lemma 5. MLE (with normal error assumption)= OLS equation-by-equation with Ω 1
P
êt ê0t
T
Intuition: all variables are included in all equations, so there is nothing gained by doing SUR. This also
implies that OLS equation by equation is asymptotically efficient.

Granger Causality
Granger Causality is a misleading name. It would be better called Granger predictability.
Definition 6. y fails to Granger cause x if it’s not helpful in linear predicting x (in MSE sense). More
formally,
h i h i
ˆ t+s |xt , ..., yt , yt−1 , ...) , ∀s > 0
MSE Ê(xt+s |xt , xt−1 , ...) = MSE E(x

ˆ t |z) denotes the best linear prediction of xt given z

where E(x
A test of Granger causality is to run OLS:
xt = α1 xt−1 + ... + αp xt−p + β1 yt−1 + ... + βp yt−p + et
and test H0 : β1 = β2 = ... = 0.
Granger causality is not economic causality. There could be simultaneous casualty, omitted variable
problems, or forward looking expectations. For example,
Example 7. Suppose there is a stock that is going to pay high dividends tomorrow. Then people will buy
the stock today, and the price will rise. We would find that prices Granger cause dividends, even though it
was really that anticipated high dividends caused high prices.

Reporting Results
Reporting the matrix of coefficients is not very informative. There are too many of them, and the coeffi-
cients are difficult to interprete anyway. Instead, people present impulse-response functions and variance
decompositions.

Impulse-response
Suppose
yt = a(L)yt−1 + et
with MA representation
yt = c(L)et
and Dut = et such that ut are orthonormal, i.e. Eut u0t = I. Let c̃(L) = c(L)D, so
yt = c̃(L)ut
 Impulse-response 3

Figure 1: Impulse-response function

Impulse Response

2.5

2
Amplitude

1.5

0.5

0 5 10 15 20 25
Samples

∂yti
Definition 8. The impulse-response function is ∂uk
. It is the change in yti in response to a unit change
t−j

in ukt−j holding all other shocks constant. We can plot the impulse-response function as in figure 1.
To estimate an impulse-response, we would
1. Estimate VAR by OLS – â
2. Invert to MA
3. Find and apply rotation D to get orthonormal shocks – the impulse response is given by ĉ˜

Standard Errors
Delta-method To calculate standard errors, we can apply the delta-method to â – the ĉ˜ are just some
complicated function of â. In practice, we can do this recursively:

yt =a1 yt−1 + ... + ap yt−p + et

=a1 (a1 yt−2 + ... + ap yt−p−1 + et−1 ) + a2 yt−2 + ... + yt−p + et
..
.

so, c1 = a1 , c2 = a2 + a12 , etc. We can apply the delta-method to each of these coefficients. We’d also need to
apply the delta-method to our estimate of D. Sometimes, this is done in practice. However, it is not really
the best way, for two reasons:
• We estimate many aj from not all that big of a sample, so our asymptotics may not be very good.
• This is made even worse by the fact that ck are highly non-linear transformations of aj
Instead of the delta-method, we can use the bootstrap.

Bootstrap
 Impulse-response 4

1
P
Example 9. ξ1 , ...ξn iid with Var(ξi ) = 1. Suppose we estimate µ̂ = n ξi and want to form a confidence
interval. We can do this by using the asymptotic distribution of µ̂,
√
n(µ̂ − µ) ⇒ N (0, 1)
1.96 1.96
µ̂ − √ ≤ µ ≤ µ̂ + √
n n

but what if we don’t think that our sample is large enough for this normal approximation to be good. How
else could we approximate the distribution of µ̂? One way would be to bootstrap:
1. From ξ1 , ...ξn , calculate µ̂ and êi = ξi − µ̂
2. For b = 1..B
(a) Form ξib = µ̂ + ebi , where ebi is sampled randomly with replacement from {êi }
(b) Calculated µ∗b = n1
P b
ξi
3. Form α confidence interval as the α/2 to 1 − α/2 quantiles of µ∗b
Remark 10. The bootstrap is still an asymptotic procedure. One advantage of the bootstrap is its simplicity.
There is no need to apply the delta-method. Another advantage is that for certain statistics (asymptotically
pivotal ones), the bootstrap is a second order refinement, i.e. it is a better asymptotic approximation than
the normal approximation.
To apply the bootstrap to an impulse-response function you could:
ˆ c̃ˆj . Also estimate êt
1. Estimate â, Ω,
2. Simulate (y1∗ , ..., yT∗ ) with â and errors resample from êt

3. Compute ĉ˜∗j , repeat many times, take quantiles of ĉ˜∗j to form confidence intervals
Remark 11. There are variations of the bootstrap that also work. For example, you could sample the errors
ˆ This would be called a parametric bootstrap because we’d be
from a normal distribution with variance Ω.
relying on a parametric assumption to create our simulated samples.