Econometrics 2 Fall 2005

Non-Stationary Time Series

and Unit Root Tests
Heino Bohn Nielsen
1 of 25
Introduction
Many economic time series are trending.
Important to distinguish between two important cases:
(1) A stationary process with a deterministic trend:
Shocks have transitory eects.
(2) A process with a stochastic trend or a unit root:
Shocks have permanent eects.
Why are unit roots important?
(1) Interesting to know if shocks have permanent or transitory eects.
(2) It is important for forecasting to know if the process has an attractor.
(3) Stationarity was required to get a LLN and a CLT to hold.
For unit root processes, many asymptotic distributions change!
Later we look at regressions involving unit root processes: spurious regression
and cointegration.
2 of 25
Outline of the Lecture
(1) Dierence between trend stationarity and unit root processes.
(2) Unit root testing.
(3) Dickey-Fuller test.
(4) Caution on deterministic terms.
(5) An alternative test (KPSS).
3 of 25
Trend Stationarity
Consider a stationary AR(1) model with a deterministic linear trend term:
Y
t
= Y
t1
+ + t +
t
, t = 1, 2, ..., T, ()
where || < 1, and Y
0
is an initial value.
The solution for Y
t
(MA-representation) has the form
Y
t
=
t
Y
0
+ +
1
t +
t
+
t1
+
2

t2
+
3

t3
+ ...
Note that the mean,
E[Y
t
] =
t
Y
0
+ +
1
t +
1
t for T ,
contains a linear trend, while the variance is constant:
V [Y
t
] = V [
t
+
t1
+
2

t2
+ ...] =
2
+
2

2
+
4

2
+ ... =

2
1
2
.
4 of 25
The original process, Y
t
, is not stationary.
The deviation from the mean,
y
t
= Y
t
E[Y
t
] = Y
t

1
t
is a stationary process. The process Y
t
is called trend-stationary.
The stochastic part of the process is stationary and shocks have transitory eects
We say that the process is mean reverting.
Also, we say that the process has an attractor, namely the mean, +
1
t.
We can analyze deviations from the mean, y
t
.
From the Frisch-Waugh theorem this is the same as a regression including a trend.
5 of 25
Shock to a Trend-Stationarity Process
0 10 20 30 40 50 60 70 80 90 100
0.0
2.5
5.0
7.5
10.0
12.5
15.0
y
t
=0.8 y
t 1
+
t
Y
t
= y
t
+ 0.1 t
6 of 25
Unit Root Processes
Consider the AR(1) model with a unit root, = 1 :
Y
t
= Y
t1
+ +
t
, t = 1, 2, ..., T, ()
or Y
t
= +
t
,
where Y
0
is the initial value.
Note that z = 1 is a root in the autoregressive polynomial, (L) = (1 L).
(L) is not invertible and Y
t
is non-stationary.
The process Y
t
is stationary. We denote Y
t
a dierence stationary process.
If Y
t
is stationary while Y
t
is not, Y
t
is called integrated of rst order, I(1).
A process is integrated of order d, I(d), if it contains d unit roots.
7 of 25
The solution for Y
t
is given by
Y
t
= Y
0
+
t
X
i=1
Y
i
= Y
0
+
t
X
i=1
( +
i
) = Y
0
+ t +
t
X
i=1

i
,
with moments
E[Y
t
] = Y
0
+ t and V [Y
t
] = t
2
Remarks:
(1) The eect of the initial value, Y
0
, stays in the process.
(2) The innovations,
t
, are accumulated to a random walk,
P

i
.
This is denoted a stochastic trend.
Note that shocks have permanent eects.
(3) The constant is accumulated to a linear trend in Y
t
.
The process in () is denoted a random walk with drift.
(4) The variance of Y
t
grows with t.
(5) The process has no attractor.
8 of 25
Shock to a Unit Root Process
0 10 20 30 40 50 60 70 80 90 100
-8
-6
-4
-2
0
2
y
t
=y
t 1
+
t
9 of 25
Unit Root Tests
A good way to think about unit root tests:
We compare two relevant models: H
0
and H
A
.
(1) What are the properties of the two models?
(2) Do they adequately describe the data?
(3) Which one is the null hypothesis?
Consider two alternative test:
(1) Dickey-Fuller test: H
0
is a unit root, H
A
is stationarity.
(2) KPSS test: H
0
is stationarity, H
A
is a unit root.
Often dicult to distinguish in practice (Unit root tests have low power).
Many economic time series are persistent, but is the root 0.95 or 1.0?
10 of 25
The Dickey-Fuller (DF) Test
Idea: Set up an autoregressive model for y
t
and test if (1) = 0.
Consider the AR(1) regression model
y
t
= y
t1
+
t
.
The unit root null hypothesis against the stationary alternative corresponds to
H
0
: = 1 against H
A
: < 1.
Alternatively, the model can be formulated as
y
t
= ( 1)y
t1
+
t
= y
t1
+
t
,
where = 1 = (1). The unit root hypothesis translates into
H
0
: = 0 against H
A
: < 0.
11 of 25
The Dickey-Fuller (DF) test is simply the t test for H
0
:
b =
b
1
se(
b
)
=
b
se(b )
.
The asymptotic distribution of b is not normal!
The distribution depends on the deterministic components.
In the simple case, the 5% critical value (one-sided) is 1.95 and not 1.65.
Remarks:
(1) The distribution only holds if the errors
t
are IID (check that!)
If autocorrelation, allow more lags.
(2) In most cases, MA components are approximated by AR lags.
The distribution for the test of (1) = 0 also holds in an ARMA model.
12 of 25
Augmented Dickey-Fuller (ADF) test
The DF test is extended to an AR(p) model. Consider an AR(3):
y
t
=
1
y
t1
+
2
y
t2
+
3
y
t3
+
t
.
A unit root in (L) = 1
1
L
2
L
2

3
L
3
corresponds to (1) = 0.
The test is most easily performed by rewriting the model:
y
t
y
t1
= (
1
1)y
t1
+
2
y
t2
+
3
y
t3
+
t
y
t
y
t1
= (
1
1)y
t1
+ (
2
+
3
)y
t2
+
3
(y
t3
y
t2
) +
t
y
t
y
t1
= (
1
+
2
+
3
1)y
t1
+ (
2
+
3
)(y
t2
y
t1
) +
3
(y
t3
y
t2
) +
t
y
t
= y
t1
+ c
1
y
t1
+ c
2
y
t2
+
t
,
where
=
1
+
2
+
3
1 = (1)
c
1
= (
2
+
3
)
c
2
=
3
.
13 of 25
The hypothesis (1) = 0 again corresponds to
H
0
: = 0 against H
A
: < 0.
The ttest for H
0
is denoted the augmented Dickey-Fuller (ADF) test.
To determine the number of lags, k, we can use the normal procedures.
General-to-specic testing: Start with k
max
and delete insignicant lags.
Estimate possible models and use information criteria.
Make sure there is no autocorrelation.
Verbeek suggests to calculate the DF test for all values of k.
This is a robustness check, but be careful!
Why would we look at tests based on inferior/misspecied models?
An alternative to the ADF test is to correct the DF test for autocorrelation.
Phillips-Perron non-parametric correction based on HAC standard errors.
Quite complicated and likely to be inferior in small samples.
14 of 25
Deterministic Terms in the DF Test
The deterministic specication is important:
We want an adequate model for the data.
The deterministic specication changes the asymptotic distribution.
If the variable has a non-zero level, consider a regression model of the form
Y
t
= Y
t1
+ c
1
y
t1
+ c
2
y
t2
+ +
t
.
The ADF test is the ttest, b
c
= b /se(b ).
The critical value at a 5% level is 2.86.
If the variable has a deterministic trend, consider a regression model of the form
Y
t
= Y
t1
+ c
1
y
t1
+ c
2
y
t2
+ + t +
t
.
The ADF test is the ttest, b
t
= b /se(b ).
The critical value at a 5% level is 3.41.
15 of 25
The DF Distributions
-5 -4 -3 -2 -1 0 1 2 3 4
0.1
0.2
0.3
0.4
0.5
N(0,1)
DF,
DF with a constant,
c
DF with a linear trend,
t
16 of 25
Empirical Example: Danish Bond Rate
1970 1975 1980 1985 1990 1995 2000 2005
-0.025
0.000
0.025
0.050
0.075
0.100
0.125
0.150
0.175
0.200
Bond rate, r
t
First difference, r
t
17 of 25
An AR(4) model gives
r
t
= 0.0093
(0.672)
r
t1
+ 0.4033
(4.49)
r
t1
0.0192
(0.199)
r
t2
0.0741
(0.817)
r
t3
+ 0.0007
(0.406)
.
Removing insignicant terms produce a model
r
t
= 0.0122
(0.911)
r
t1
+ 0.3916
(4.70)
r
t1
+ 0.0011
(0.647)
.
The 5% critical value (T = 100) is 2.89, so we do not reject the null of a unit root.
We can also test for a unit root in the rst dierence.
Deleting insignicant terms we nd a preferred model

2
r
t
= 0.6193
(7.49)
r
t1
0.00033
(0.534)
.
Here we safely reject the null hypothesis of a unit root (7.49 2.89).
Based on the test we concluce that r
t
is non-stationary while r
t
is stationary.
That is r
t
I(1)
18 of 25
A Note of Caution on Deterministic Terms
The way to think about the inclusion of deterministic terms is via the factor repre-
sentation:
y
t
= y
t1
+
t
Y
t
= y
t
+
It follows that
(Y
t
) = (Y
t1
) +
t
Y
t
= Y
t1
+ (1 ) +
t
Y
t
= Y
t1
+ +
t
which implies a common factor restriction.
If = 1, then implicitly the constant should also be zero, i.e.
= (1 ) = 0.
19 of 25
The common factor is not imposed by the normal ttest. Consider
Y
t
= Y
t1
+ +
t
.
The hypotheses
H
0
: = 1 against H
A
: < 1,
imply
H
A
: Y
t
= + stationary process
H
0
: Y
t
= Y
0
+ t + stochastic trend.
We compare a model with a linear trend against a model with a non-zero level!
Potentially dicult to interpret.
20 of 25
A simple alternative is to consider the combined hypothesis
H

0
: = = 0.
The hypothesis H

0
can be tested by running the two regressions
H
A
: Y
t
= Y
t1
+ +
t
H

0
: Y
t
=
t
,
and perform a likelihood ratio test

LR
= T log

RSS
0
RSS
A

= 2 (loglik
0
loglik
A
) ,
where RSS
0
and RSS
A
denote the residual sum of squares.
The 5% critical value is 9.13.
21 of 25
Same Point with a Trend
The same point could be made with a trend term
Y
t
= Y
t1
+ + t +
t
.
Here, the common factor restriction implies that if = 0 then = 0.
Since we do not impose the restriction under the null, the trend will accumulate.
A quadratic trend is allowed under H
0
, but only a linear trend under H
A
.
A solution is to impose the combined hypothesis
H

0
: = = 0.
This is done by running the two regressions
H
A
: Y
t
= Y
t1
+ + t +
t
H

0
: Y
t
= +
t
,
and perform a likelihood ratio test. The 5% critical value for this test is 12.39.
22 of 25
Special Events
Unit root tests assess whether shocks have transitory or permanent eects.
The conclusions are sensitive to a few large shocks.
Consider a one-time change in the mean of the series, a so-called break.
This is one large shock with a permanent eect.
Even if the series is stationary, such that normal shocks have transitory eects, the
presence of a break will make it look like the shocks have permanent eects.
That may bias the conclusion towards a unit root.
Consider a few large outliers, i.e. a single strange observations.
The series may look more mean reverting than it actually is.
That may bias the results towards stationarity.
23 of 25
A Reversed Test: KPSS
Sometimes it is convenient to have stationarity as the null hypothesis.
KPSS (Kwiatkowski, Phillips, Schmidt, and Shin) Test.
Assume there is no trend. The point of departure is a DGP of the form
Y
t
=
t
+ e
t
,
where e
t
is stationary and
t
is a random walk, i.e.

t
=
t1
+ v
t
, v
t
IID(0,
2
v
).
If the variance is zero,
2
v
= 0, then
t
=
0
for all t and Y
t
is stationary.
Use a simple regression:
Y
t
= b + b e
t
, ()
to nd the estimated stochastic component. Under the null, b e
t
is stationary.
That observation can be used to design a test:
H
0
:
2
v
= 0 against H
A
:
2
v
> 0.
24 of 25
The test statistic is given by
KPSS =
1
T
2

P
T
t=1
S
2
t
b
2

,
where S
t
=
P
t
s=1
b e
s
is a partial sum; b
2

is a HAC estimator of the variance of b e

t
.
(This is an LM test for constant parameters against a RW parameter).
The regression in () can be augmented with a linear trend. Critical values:
Deterministic terms Critical values
in regression () 0.10 0.05 0.01
Constant 0.347 0.463 0.739
Constant and trend 0.119 0.146 0.216
Can be used for conrmatory analysis:
KPSS:
Rejection of I(0) Non-rejection of I(0)
DF: Rejection of I(1) ? I(0)
Non-rejection of I(1) I(1) ?
25 of 25