BEE2006 Statistics and Econometrics
Tutorial 3: Further Issues in using OLS with Time series data
February 2013
Question 11.1
Let {xt:t = 1,2, ....} be a covariance stationary process define γh=Cov(xt,xt+h) for h≥0
Show that
Corr(xt,xt+h) = γh
γ0
Main Points:
A stochastic process {xt:t = 1,2, ....} is covariance stationary if and only if
1 E(xt) =µfor allt = 1,2, ...
2 Var(xt) =σ2 for allt= 1,2, ...
3 γh=Cov(xt,xt+h) for h≥0 andt = 1,2,3, ...
theCov(xt,xt+h) is independent oft
Why Stationary is important:
1 Simplifies the central limit theorem
2 If we want to study the relationship between yt andxt, we want to make “say something” that will be robust across time
Recall that:
Corr(xt,xt+h) = Cov(xt,xt+h)
!Var(xt)Var(xt+h)
by definition Cov(xt,xt+h) =γh
Var(xt) =Cov(xt,xt) =γ0 Var(xt+h) =Cov(xt+h,xt+h) =γ0
Therefore:
Corr(xt,xt+h) = √VarCov(xt,xt+h)
(xt)Var(xt+h)
= √γh γ0γ0
= γh
γ0
Question 11.2 (i)
Let {et :t =−1,0,1,2, ...} be a sequence of independent, identically distributed random variables with mean zero and variance one.
Define a stochastic process by
xt =et−1
2et−1+ 1
2et−2 for t = 1,2, ...
Find E(xt) and Var(xt)
From the description of et, we have thatE(et) = 0:
E(xt) = E(et)−12E(et−1) +12E(et−2)
= 0 + 0 + 0
= 0
Most importantly we have thatE(xt) is independent of time
From the description of et, Var(et) = 1
Cov(et,es) = 0 for allt $=s since by the iid assumptions Var(xt) = Var(et)−14Var(et−1) + 14Var(et−2)
= 1 +14 +14
= 32
Most importantly we have thatVar(xt) is independent of time
Question 11.2 (ii-iii)
Show that Corr(xt,xt+1) =−1/2 andCorr(xt,xt+2) = 1/3 Find the general expression for Corr(xt,xt+h) for h>2
Firstly
Corr(xt,xt+1) = Cov(xt,xt+1)
!Var(xt)Var(xt+1)
Var(xt) =Var(xt+1) = 32 Cov(xt,xt+1) =Cov
"
et−1
2et−1+1
2et−2,et+1− 1 2et+1
2et−1
#
Cov(xt,xt+1) =Cov
"
et,−1 2et
# +Cov
"
−1 2et−1,1
2et−1
#
Cov(xt,xt+1) =−1
2(1)−1
4(1) =−3 4
Therefore
Corr(xt,xt+1) = Cov(xt,xt+1)
!Var(xt)Var(xt+1) = −34
$3 2
%3
2
&
=−1 2
Again
Corr(xt,xt+1) = Cov(xt,xt+2)
!Var(xt)Var(xt+2)
Var(xt) =Var(xt+2) = 32 Cov(xt,xt+2) =Cov
"
et−1
2et−1+1
2et−2,et+2− 1
2et+1+1 2et
#
Cov(xt,xt+2) =Cov
"
et,1 2et
#
Cov(xt,xt+2) = 1 2(1)
Therefore
Corr(xt,xt+2) = Cov(xt,xt+2)
!Var(xt)Var(xt+2) =
1 2
$3 2
%3
2
&
= 1 3
In General
Corr(xt,xt+h) = Cov(xt,xt+h)
!Var(xt)Var(xt+h) = 0 where h>2
Var(xt) =Var(xt+h) = 32 Cov(xt,xt+h) =Cov
"
et−1
2et−1+1
2et−2,et+h−1
2et+h−1+1 2et+h−2
#
Cov(xt,xt+h) = 0
Question 11.2 (iv)
Is {xt}an asymptotically uncorrelated process?
Notice that:
Corr(xt,xt+1)$= 0 Corr(xt,xt+2)$= 0 Corr(xt,xt+3) = 0 ...
Corr(xt,xt+h) = 0 for allh >2
So it’s obvious that Corr(xt,xt+s)→0 ass → ∞. Therefore {xt} is an asymptotically uncorrelated process.
The{xt} process is weakly dependent
Question 11.3 (i)
Suppose that a time series process {yt} is generated by yt =z+et ∀t= 1,2, ....
where {et} is aiid sequence with mean zero and variance σe2. The random variablez does not change over time; it has a mean of zero and variance of σz2. Assume that eachet is uncorrelated with z
Find the expected value and variance of yt.
Quick recap
E(et) = 0, Var(et) =σe2 andCov(et,es) = 0 for all t$=s E(z) = 0 and Var(z) =σz2
Corr(et,z) = 0 for allt = 1,2,3, ....
Therefore
E(yt) =E(z) +E(et) = 0 furthermore
Var(yt) =Var(z) +Var(et) + 2Cov(et,z) =σe2+σ2z
Question 11.3 (ii)
Is {yt}covariance stationary?
Recall that this requires
E(yt) , Var(yt) does not depend ont
Cov(yt,yt+h) forh≥1, depends onh and not ont
Now we only need to find Cov(yt,yt+h) for h≥1 Whenh = 1 then
Cov(yt,yt+1) =Cov(z+et,z+et+1) =Cov(z,z) =σz2
Whenh = 2 then
Cov(yt,yt+2) =Cov(z+et,z+et+2) =Cov(z,z) =σz2
Whenh ≥1 then
Cov(yt,yt+h) =Cov(z+et,z+et+h) =Cov(z,z) =σ2z
Therefore {yt} is covariance stationary
Question 11.3 (iii)
Show that
Corr(yt,yt+h) = σ2z
σz2+σe2
for allt andh
Notice that the key work here is “for all t and h”it doesn’t make sense to consider the correlation between Corr(yt,yt). Therefore once agains we consider for all h≥1
Corr(yt,yt+h) = Cov(yt,yt+h)
!Var(yt)Var(yt+h) = σ2z
σz2+σe2
Question 11.3 (iv)
Does the {yt}satisfy the intuitive requirement for being asymptotically uncorrelated?
is {yt}a weakly dependent process?
Recall that:
Corr(yt,yt+h) = σ2z
σz2+σe2
but
Corr(yt,yt+h)→ σ2z
σz2+σe2
as h→ ∞
therefore {yt}is not a weakly dependent process.
Question 10.6
Suppose yt follows a second order FDL model:
yt =α0+δ0zt+δ1zt−1+δ2zt−2+ut
let z∗ denote the equilibrium value ofzt andy∗ be the equilibrium value of yt such that
y∗ =α0+δ0z∗+δ1z∗+δ2z∗+ut
show that the change in y∗, due to a change inz∗, equals the long-run propensity times the change in z∗
∆y∗=LRP×∆(z∗)
The economy is at equilibrium such that yt =y∗ andzt =z∗ If there is no change in z∗ then we have it that:
y∗=α0+δ0z∗+δ1z∗+δ2z∗+ut
The economy is at equilibrium such that yt =y∗ andzt =z∗ Assume that at timet there is a permanent change inz∗ by∆ yt=α0+δ0(z∗+∆) +δ1z∗+δ2z∗+ut
yt+1 =α0+δ0(z∗+∆) +δ1(z∗+∆) +δ2(z∗) +ut
yt+2 =α0+δ0(z∗+∆) +δ1(z∗+∆) +δ2(z∗+∆) +ut
...
yt+h=α0+δ0(z∗+∆) +δ1(z∗+∆) +δ2(z∗+∆) +ut for h ≥2
Therefore in the longrun:
ynew∗ =α0+(δ0+δ1+δ2)∆+(δ0+δ1+δ2)z∗+ut =y∗+(δ0+δ1+δ2)∆ intuitively the long run equilibrium changes from y∗ to
y∗+ (δ0+δ1+δ2)∆ therefore
∆y∗ = (δ0+δ1+δ2)∆z∗