Unit Root Tests

Mei-Yuan Chen
Department of Finance
National Chung Hsing University

February 25, 2013

M.-Y. Chen I(0) vs I(1)

Unit Root Tests

An ARMA(p,q) (autoregressive of order p and moving average

of order q) model of yt is typically written as
Φ(B)yt = Ψ(B)ǫt , where

Φ(B) = 1 − φ1 B − φ2 B 2 − · · · − φp B p ,
Ψ(B) = 1 − ψ1 B − ψ2 B 2 − · · · − ψq B q ,

are polynomials in B, the back-shift operator, and {ǫt } is

white-noise.

M.-Y. Chen I(0) vs I(1)

Stationarity

An ARMA model is said to be “stationary” if all the roots of

Φ(z) = 0 are outside the unit circle, |z| = 1. In this case, we
can obtain a stationary solution yt = Φ(B)−1 Ψ(B)ǫt .

M.-Y. Chen I(0) vs I(1)

Invertibility

An ARMA model is said to be “invertible” if all the roots of

Ψ(z) = 0 are outside the unit circle. In this case, we can write
ǫt = Ψ(B)−1 Φ(B)yt . For example, if yt = αyt−1 + ǫt , then
{yt } is stationary provided that the root of Φ(z) = 1 − αz = 0,
z = 1/α, is outside the unit circle, i.e., |α| < 1.

M.-Y. Chen I(0) vs I(1)

ARIMA Models

An ARIMA(p,d,q) model of yt is Φ(B)(l − B)d yt = Ψ(B)ǫt ,

where Φ(B) and Ψ(B) satisfy the stationarity and invertibility
conditions. As the polynomial A(z) ≡ Φ(z)(1 − z)d = 0 has d
roots on the unit circle, {yt } is said to be an I(d) process
(integrated process of order d). That is, an I(d) process must
be differenced d times to achieve stationarity; in particular, an
I(0) process is stationary.

M.-Y. Chen I(0) vs I(1)

Of particular interest to us are I(1) processes of the following
form: (1 − B)yt = ut , where ut has a stationary and invertible
ARMA representation. To see the properties of an I(1)
process, consider a special case where {yt } is a “random
walk”, i.e., the innovations ut are i.i.d. with mean zero and
finite variance σu2 . That is,

yt = yt−1 + ut , ut ∼ i.i.d.(0, σu2 ).

M.-Y. Chen I(0) vs I(1)

Properties of a Random Walk

1. the effects of past ut on yt are permanent;

2. yt has unbounded variance tσu2 ;
3. yt has a smooth sample path (in terms of level crossing)
which typically wanders away and rarely crosses its mean
level. By contrast, an I(0) process has short memory,
bounded variance, and a ragged sample path which
crosses its mean level very often.

M.-Y. Chen I(0) vs I(1)

Clearly, an I(1) process is non-stationary, but a non-stationary
series need not be I(1). If (1 − B)yt has a non-zero mean µ0
such that (1 − B)yt = µ0 + ut , then {yt } is said to be an I(1)
process with the drift µ0 (since yt = µ0 t + ti=0 ut−i . A major
P

problem in time-series regressions of an I(1) variable is that

many known results are no longer valid.

M.-Y. Chen I(0) vs I(1)

I(1) without Drift

Consider the DGP (data generating process):

yt = α0 yt−1 + ut ,

where ut is an I(0) process. We want to test the null

hypothesis of a unit root (α0 = 1) against the alternative
hypothesis that yt is I(0) (|α0 | < 1).

M.-Y. Chen I(0) vs I(1)

Dickey-Fuller Tests

Suppose that we estimate one of the following three models.

yt = α̂1T yt−1 + ût , (1)

yt = µ̂2T + α̂2T yt−1 + ût , (2)
 T
yt = µ̂3T + β̂3T t − + α̂3T yt−1 + ût , (3)
2
where ût is a generic notation for OLS residual.

M.-Y. Chen I(0) vs I(1)

If ut is i.i.d. N(0, σu2 ), Dickey & Fuller (1979) show that under
the null hypothesis T (α̂iT − 1) and the t-statistics

τα̂iT = (α̂iT − 1)/[s.e.(α̂iT )], i = 1, 2, 3,

τµ̂iT = µ̂iT /[s.e.(µ̂iT )], i = 2, 3,
τβ̂3T = β̂3T /[s.e.(β̂3T )],

have non-normal limiting distributions, where s.e. stands for

the OLS standard errors.

M.-Y. Chen I(0) vs I(1)

The empirical distributions of T (α̂iT − 1) and τα̂iT , i = 1, 2, 3
are tabulated in Fuller (1976, p. 371 & p. 373); the
distributions of the t-ratios τµ̂iT , i = 2, 3 and τβ̂3T are
tabulated in Dickey & Fuller (1981). From the table for
T (α̂1T − 1) it can be seen that this distribution is skewed to
the left and α̂1T is downward biased.

M.-Y. Chen I(0) vs I(1)

Table Finite-sample and asymptotic critical values of the
Dickey-Fuller test.

Sample Significance Level

Size 1% 2.5% 5% 10%
25 −3.75 −3.33 −3.00 −2.63
50 −3.58 −3.22 −2.93 −2.60
100 −3.51 −3.17 −2.89 −2.58
250 −3.46 −3.14 −2.88 −2.57
∞ −3.43 −3.12 −2.86 −2.57

M.-Y. Chen I(0) vs I(1)

The distributions of T (α̂iT − α0 ) when α0 = −1 are just the
mirror images of those for α0 = 1, in the sense that

P {α̂iT − α0 > c|α0 = 1} = P {α̂iT − α0 < −c|α0 = −1}.

Homework! Simulate and plot the empirical distributions of

α̂iT under the DGPs (data generating process) with α0 = 1
and α0 = −1 for T = 100 and replications 2000.

M.-Y. Chen I(0) vs I(1)

Augmented Dickey-Fuller Tests
More generally, suppose that ut has an ARMA(p, q)
representation: Φ(B)ut = Ψ(B)ǫt , where ǫt are i.i.d. N(0, σ 2 ).
First consider the case that p = q = 1. Then
ut − φ1 ut−1 = ǫt − ψ1 ǫt−1 , and
ǫt
= ut − φ1 ut−1 + ψ1 ǫt−1
= (ut − φ1 ut−1 ) + ψ1 (ut−1 − φ1 ut−2 + ψ1 ǫt−2 )
= (ut − φ1 ut−1 ) + ψ1 (ut−1 − φ1 ut−2 ) + ψ12 (ut−2 − φ1 ut−3 + ψ1 ǫt−3 )
.
= ..
∞
ψ1j (ut−j − φ1 ut−1−j ).
X
=
j=0

M.-Y. Chen I(0) vs I(1)

 ψ1j−1(ut−j − φ1 ut−1−j ),
P∞
We have, since ǫt−1 = j=1

yt − yt−1
= (α0 yt−1 + ut ) − yt−1
= (α0 − 1)yt−1 + φ1 ut−1 − ψ1 ǫt−1 + ǫt
∞
ψ1j−1 (ut−j − φ1 ut−1−j ) + ǫt
X
= (α0 − 1)yt−1 + φ1 ut−1 − ψ1
j=1

= (α0 − 1)yt−1 + (φ1 − ψ1 )[ut−1 + ψ1 ut−2 + ψ12 ut−3 + · · · ] + ǫt .

M.-Y. Chen I(0) vs I(1)

This suggests a regression of ∆yt on yt−1 and ∆yt−1 , ∆yt−2 ,
. . . , which can be approximated using an autoregression of
finite order k:

∆yt ≈ (α0 − 1)yt−1 +

(φ1 − π1 )[∆yt−1 + ψ1 ∆yt−2 + ψ12 ∆yt−3 + · · ·
+ψ1k−1 ∆yt−k ] + ǫt
=: θ0 yt−1 + θ1 ∆yt−1 + θ2 ∆yt−2 + · · · + θk ∆yt−k + ǫt .

M.-Y. Chen I(0) vs I(1)

To obtain consistent estimates, the order of autoregression, k,
must be a function of T ; in particular, Said & Dickey (1984)
show that k must be o(T 1/3 ) and that there exist c > 0 and
r > 0 such that ck > T 1/r .

M.-Y. Chen I(0) vs I(1)

On the other hand, the t-statistic for θ̂0T has the same limiting
distribution as that of τα̂1T ; hence the table in Fuller (1976)
can be used. If a constant term is included in the model, the
resulting t-statistic has the same limiting distribution as that
of τα̂2T . The same idea carries over if ut follows an
ARMA(p, q) model. Tests of this type are known as the
Augmented Dickey-Fuller (ADF) tests.

M.-Y. Chen I(0) vs I(1)

Phillips-Perron Tests

More general results are available when ut are weakly

dependent. Let
" T # T
1  X 2 1X
σ 2 := lim E ut , σu2 := lim E(ut )2 .
T →∞ T T →∞ T
t=1 t=1

M.-Y. Chen I(0) vs I(1)

Define YT in D[0, 1] as
[T r]
1 X
YT (r) = √ ut ,
σ T t=1

where [T r] denotes the integer part of T r. We assume

suitable conditions under which the FCLT holds, i.e.,
YT ⇒ W , where W is a standard Wiener process. More
specific conditions can be found in, e.g., Phillips (1987) and
Wooldridge & White (1988).

M.-Y. Chen I(0) vs I(1)

The results below are fundamental for Phillips (1987) and
many other papers written by Phillips and his co-authors.
Without loss of generality we set y0 = 0. Let
R1
W ∗ = W − 0 W (r) dr. If yt = ti=1 ui ,
P
Z 1
PT
(1) T −3/2
t=1 yt−1 ⇒ σ W (r) dr;
0
Z 1
PT 2 2
(2) T −2
t=1 yt−1 ⇒ σ W (r)2 dr;
0

M.-Y. Chen I(0) vs I(1)

 PT
(3) T −1 t=1 yt−1 ut ⇒
1
1 1
Z
2
σ W (r) dW (r) + (σ 2 − σu2 ) = (σ 2 W (1)2 − σu2 );
0 2 2
P T
(4) T −3/2 t=1 tut ⇒ 
Z 1 Z 1 
σ r dW (r) = σ W (1) − W (r) dr ;
0 0
Z 1
PT
(5) T −5/2
t=1 tyt−1 ⇒ σ rW (r) dr;
0
Z 1
PT 2
(6) T −2 2
t=1 (yt−1 − ȳ−1 ) ⇒ σ W ∗ (r)2 dr;
0
PT
(7) T −1
(y − ȳ−1 )ut ⇒
Z 1 t=1 t−1
1
σ2 W ∗ (r) dW (r) + (σ 2 − σu2 ) .
0 2

M.-Y. Chen I(0) vs I(1)

Under the null hypothesis that yt = yt−1 = ut , where {ut }
satisfies the FCLT, then
(a) For model (1),
1


2
W (1)2 − σu2 /σ 2
T (α̂1T − 1) ⇒ R1 ;
W (r) 2 dr
0
σ


2σu
W (1)2 − σu2 /σ 2
τα̂1T ⇒ R 1/2 .
1 2
0
W (r) dr

M.-Y. Chen I(0) vs I(1)

(b) Let λ = 12 (σ 2 − σu2 ) and λ∗ = λ/σ 2 . For model (2),
R1
0
W ∗ (r) dW (r) + λ∗
T (α̂2T − 1) ⇒ R1 ;
0
W ∗ (r)2 dr

and

M.-Y. Chen I(0) vs I(1)

 R 
σ 1 ∗ ∗
σu 0
W (r) dW (r) + λ
τα̂2T ⇒ R 1/2 ;
1 2
0
W (r) dr
∗
hR  R  R i
σ 1 2 1 1 ∗
σu 0 W (r) dr W (1) − 0 W (r) dr 0 W (r) dW (r) + λ
τµ̂2T ⇒ R 1/2 R 1/2
1 ∗ (r)2 dr 1 2 dr
0 W 0 W (r)
 R 2 
1 ∗ ∗
2
σ  0 W (r) dW (r) + λ
ϕ1 ⇒ + W (1)2  .

R1
2σu2

W (r) dr
∗ 2
0

M.-Y. Chen I(0) vs I(1)

Remarks
(1) For model (3), the limits of T (α̂3T − 1), τµ̂3T , τβ̂3T , and
τα̂3T can be found in Phillips & Perron (1988), and the
limits of φ2 and φ3 can be found in Perron (1990).
(2) The OLS estimators α̂iT , i = 1, 2, 3, are all ”supper
consistent” in the sense that α̂iT − 1 is Op (T −1 ), whereas
in the standard case such that |α0 | < 1, αiT − α0 is
Op (T −1/2 ). Moreover, the OLS estimators remain
consistent even when yt−1 and serially correlated
disturbances are both present because the sample
variation Tt=1 yt−1
2
P
grows much faster than the
regressor-error correlation Tt=1 yt−1 ut .
P

M.-Y. Chen I(0) vs I(1)

 PT
Note that under the null hypothesis, the estimators σ̂T2 = T −1 t=1 û2t
T
and T −1 t=1 (yt − yt−1 )2 are consistent for σu2 by suitable law of large
P

numbers. Also observe that σ 2 is the limit of

" T # T T −1 X T
!
1  X 2 1 X X
2
E ut = E(ut ) + 2 E(ut ut−τ ) ,
T t=1
T t=1 τ =1 t=τ +1

which can be consistently estimated using the following (non-parametric)

estimator
 
T n(T ) T
1 X X X
s2T,n(T ) = û2t + 2 wτ n ût ût−τ  ,
T
t=1 τ =1 t=τ +1

where w(·) is some kernel.

M.-Y. Chen I(0) vs I(1)

Note that n(T ) characterizes kernel’s bandwidth which should be
growing with T but at a slower rate; especially, n(T ) can be o(T 1/2 ). If
we use the so-called Bartlett kernel:
(
1 − τ /(n(T ) + 1), if 0 ≤ τ /n(T ) ≤ 1,
wτ n =
0, otherwise,

we obtain the Newey & West (1987) estimator; if we use the so-called
Parzen kernel:

2 3
 1 − 6[τ /n(T )] + 6[τ /n(T )] , if 0 ≤ τ /n(T ) ≤ 1/2,

wτ n = 3
2[1 − τ /n(T )] , if1/2 ≤ τ /n(T ) ≤ 1,

0, otherwise,


we obtain the Gallant (1987) estimator.

M.-Y. Chen I(0) vs I(1)

Limiting Distributions of D-F Tests
If ut are i.i.d. with mean zero and variance σ 2 , then σu2 = σ 2
and λ = λ∗ = 0. Under the null hypothesis that
yt = yt−1 + ut , where {ut } satisfies the FCLT, then:
(a) For model (1),
1


2
W (1)2 − 1
T (α̂1T − 1) ⇒ R 1 ;
0
W (r)2 dr
1 2


W (1) − 1
τα̂1T ⇒ R2 1/2 .
1 2
0
W (r) dr

M.-Y. Chen I(0) vs I(1)

(b) for model (2),
R1
0
W ∗ (r) dW (r)
T (α̂2T − 1) ⇒ R1 ;
0
W ∗ (r)2 dr
R 
1 ∗
0
W (r) dW (r)
τα̂2T ⇒ R 1/2 ;
1 ∗ (r)2 dr
0
W
R  R  R 
1 2 1 1
0 W (r) dr W (1) − 0 W (r) dr 0 W (r) dW (r)
τµ̂2T ⇒ R 1/2 R 1/2
1 1
0
W ∗ (r)2 dr 0
W (r)2 dr
 R 2 
1 ∗
1 0 W (r) dW (r)
ϕ1 ⇒ + W (1)2  .

R1
2

0
W ∗ (r)2 dr

M.-Y. Chen I(0) vs I(1)

PP Z-Test

As the limits of the DF tests depend on nuisance parameters

when ut are weakly dependent, critical values cannot be
tabulated. Fortunately, the nuisance parameters can be
eliminated even when ut are not i.i.d.

M.-Y. Chen I(0) vs I(1)

Consider
1 2 2
2 (sT n − σ̂T )
Z(α̂1T ) = T (α̂1T − 1) − T 2
P
T −2 t=1 yt−1
1 2 2 2 1 2 2
2 (W (1) − σu /σ ) 2 (σ − σu )
⇒ R1 − R1
2 σ 2 0 W (r)2 dr
0 W (r) dr
1 2
2 (W (1) − 1)
= R1 ,
0
W (r)2 dr

which is the same as the limit of T (α̂1T − 1). Therefore, the table in
Fuller (1976) can be directly used for this test. A modified
√
Zα′ = Z(α̂1T )/ 2 is related to the empirical tables in Evans &
Savin (1981).

M.-Y. Chen I(0) vs I(1)

Similarly,
1 2
σ̂T 2 (sT n − σ̂T2 )
Z(τα̂1T ) = τα̂ − 1/2
sT n 1T  PT 2
sT n T −2 t=1 yt−1
1 2 1
2 (W (1) − σu2 /σ 2 ) 2 2
2 (σ − σu )
⇒ R 1/2 − R 1/2
1 1
0 W (r)2 dr σ 2 0 W (r)2 dr
1 2
2 (W (1) − 1)
= R 1/2 .
1
0
W (r)2 dr

M.-Y. Chen I(0) vs I(1)

Dickey-Fuller Tests with GLS Detrending Data

ERS (1996) propose a simple modification of the ADF tests in

which the data are detrended so that explanatory variables are
”taken out” of the data prior to running the test regression.
ERS define a quasi-difference of that depends on the value
representing the specific point alternative against which we
wish to test the null:
( (
yt if t = 1, xt if t = 1,
d(yt |a) = , d(xt |a) =
yt − ayt−1 if t < 1, xt − ayt−1 if t < 1,

M.-Y. Chen I(0) vs I(1)

Next, consider an OLS regression of the quasi-differenced data
d(yt |a) on the quasi-differenced d(xt |a):

d(yt |a) = d(xt |a)′ δ(a) + ut

where xt contains either a constant, or a constant and trend,

and let δ̂(a) be the OLS estimates from this regression.

M.-Y. Chen I(0) vs I(1)

Now, all that we need know is a value for a. ERS recommend
the use of ā, where:
(
1 − T7 if xt = 1,
ā = 13.5
1 − T if xt = (1, t)′ .

Then, define the GLS detrended data, ytd using the estimates
associated with the ā:

ytd = yt − x′t δ̂(ā).

M.-Y. Chen I(0) vs I(1)

Then the DFGLS test involves estimating the standard ADF
test equation after substituting the GLS detrended ytd for the
original yt :
Xp
d d d
△yt = αyt−1 + βj △yt−j + vt .
j=1

Note that since the ytd are detrended, we do not include the xt
in the DFGLS test equation. As with the ADF test, we
consider the t-ratio for α from this test equation.
While the DFGLS -ratio follows a Dickey-Fuller distribution in
the constant only case, the asymptotic distribution differs
when you include both a constant and trend. ERS (1996,
Table 1, p. 825) simulate the critical values of the test
statistic in this latter setting for T = {50, 100, 200, . . . , ∞}.
M.-Y. Chen I(0) vs I(1)
Homework! Simulate the empirical sizes and powers of ADF
and DFGLS tests under the DGP: yt = 0.1 + α0 yt−1 + ut with
α0 = 1 and α0 = 0.8. The sample size and replications are
considered as T = 100 and 2000.

M.-Y. Chen I(0) vs I(1)

Unit Root Tests in gretl

The unit root tests of ADF and ADF-GLS are provided in

gretl. These two tests are accessed via 「Variable (V)」 in tool
bar of gretl.

M.-Y. Chen I(0) vs I(1)

Unit Root Tests in R

The following packages in R provide unit root tests:

1. urca; see twi− unit.R
2. tseries
3. fUnitRoots
In urca, the unit root tests can be implemented by using
urTest(x, method = c(”unitroot”, ”adf”, ”urers”, ”urkpss”, ”urpp”,
”ursp”, ”urza”), title = NULL, description = NULL, ...)

M.-Y. Chen I(0) vs I(1)

Unit Root Tests in urca
1. ur.df: Augmented-Dickey-Fuller Unit Root Test
ur.df(y, type = c(”none”, ”drift”, ”trend”), lags = 1, selectlags =
c(”Fixed”, ”AIC”, ”BIC”))
2. ur.ers: Elliott, Rothenberg & Stock Unit Root Test
ur.ers(y, type = c(”DF-GLS”, ”P-test”), model = c(”constant”,
”trend”), lag.max = 4)
3. ur.kpss: Kwiatkowski et al. Unit Root Test
ur.kpss(y, type = c(”mu”, ”tau”), lags = c(”short”, ”long”, ”nil”),
use.lag = NULL)
4. ur.pp: Phillips & Perron Unit Root Test
ur.pp(x, type = c(”Z-alpha”, ”Z-tau”), model = c(”constant”,
”trend”), lags = c(”short”, ”long”), use.lag = NULL)
M.-Y. Chen I(0) vs I(1)
Unit Root Tests in urca: continue

5. ur.sp: Schmidt & Phillips Unit Root Test

ur.sp(y, type = c(”tau”, ”rho”), pol.deg = c(1, 2, 3, 4), signif =
c(0.01, 0.05, 0.1))

6. ur.za: Zivot & Andrews Unit Root Test

ur.za(y, model = c(”intercept”, ”trend”, ”both”), lag=NULL)

M.-Y. Chen I(0) vs I(1)

Unit Root Tests in tseries

1. adf.test: Augmented Dickey-Fuller Test

adf.test(x, alternative = c(”stationary”, ”explosive”), k =
trunc((length(x) − 1)( 1/3)))

2. kpss.test: KPSS Test for Stationarity

kpss.test(x, null = c(”Level”, ”Trend”), lshort = TRUE)

3. pp.test: Phillips-Perron Unit Root Test

pp.test(x, alternative = c(”stationary”, ”explosive”), type =
c(”Z(α)”, ”Z(tα )”), lshort = TRUE)

M.-Y. Chen I(0) vs I(1)

Unit Root Tests in fUnitRoots
1. adfTest: Augmented Dickey-Fuller test for unit roots,
adfTest(x, lags = 1, type = c(”nc”, ”c”, ”ct”), title = NULL,
description = NULL)
2. unitrootTest: the same based on McKinnons’s test statistics.
unitrootTest(x, lags = 1, type = c(”nc”, ”c”, ”ct”), title = NULL,
description = NULL)
3. urdfTest: Augmented Dickey-Fuller test for unit roots,
4. urersTest: Elliott-Rothenberg-Stock test for unit roots,
5. urkpssTest: KPSS unit root test for stationarity,
6. urppTest: Phillips-Perron test for unit roots,
7. urspTest: Schmidt-Phillips test for unit roots,
8. urzaTest: Zivot-Andrews test for unit roots.
M.-Y. Chen I(0) vs I(1)
Testing Stationarity Against a Unit Root

We have learned that the DF tests usually have low power

against trend stationarity. To complement the aforementioned
unit-root tests, it is quite natural to consider tests from an
opposite direction, i.e., tests of the null hypothesis of (trend)
stationarity against a unit root.

M.-Y. Chen I(0) vs I(1)

KPSS Tests
Consider a linear regression model

yt = xt at + zt′ b0 + ǫt , t = 1, · · · , T,

where at = at−1 + vt with the initial value a0 , and vt are i.i.d.

N(0, σv2 ) independent of ǫt which are also i.i.d. N(0, σǫ2 ).
Under the null that at = a0 , σv2 = 0. Under the alternative of
random walk,
t
X
yt = xt a0 + zt′ b0 + xt vj + ǫt , t = 1, · · · , T.
j=1

M.-Y. Chen I(0) vs I(1)

Nabeya & Tanaka (1988) derive the locally best invariant test
for σv2 = 0, which is based on the ratio of two quadratic forms:

e′ Dx AT Dx e
,
e′ e
where Dx = diag(x1 , . . . , xT ), AT is such that its (s, t)th
element is min(s, t), and e is the residual vector from
regressing yt on xt and zt .

M.-Y. Chen I(0) vs I(1)

Kwiatkowski, Phillips, Schmidt, & Shin (1992) apply this test
to test for the null of (trend) stationarity. Consider the DGP:

yt = a0 + b0 t + ǫt ,

which is a special case of the null model of Nabeya &

Tanaka (1988) with xt = 1 and zt = t; when b0 = 0, this is
just a level-stationary model. Under the alternative that yt is
an I(1) process with drift,

yt = a0 + b0 t + yt∗,
Pt
where yt∗ = j=1 vj .

M.-Y. Chen I(0) vs I(1)

Note that AT can be written as CT′ CT , where CT is an upper
triangular matrix with all elements on and above the main
diagonal being 1. In this case, Dx = IT so that
e′ Dx AT Dx e = e′ CT′ CT e, and it can be verified that CT e is a
vector containing reversed partial sums of et , i.e., the t-th
element is Rt = Ti=t ei . Let St denote partial sums of et .
P

Then, R1 = ST = 0 and St = −Rt+1 for t = 1, . . . , T − 1.

M.-Y. Chen I(0) vs I(1)

Nabeya and Tanaka’s statistics can be written as:

T −2 Tt=1 Rt2 T −2 Tt=1 St2

P P
ηi = = , i = 1, 2,
σ̂T2 σ̂T2

with σ̂T2 = e′ e/T , where e is obtained from regressing yt on

the constant 1 for the null of level stationarity (i = 1) and
from regressing yt on 1 and t for the null of trend stationarity
(i = 2).

M.-Y. Chen I(0) vs I(1)

To allow for weakly dependent and heterogeneous ǫt , the
statistics become

T −2 Tt=1 St2
P
ηi = , i = 1, 2,
s2T n

where s2T n is the Newey-West or Gallant estimator of

σ 2 = limT T −1 IE(ST2 ); these two tests will be referred to as
the KPSS tests.

M.-Y. Chen I(0) vs I(1)

Theorem: KPSS

In the level-stationary model,

Z 1
η1 ⇒ W 0 (r)2 dr,
0

where W 0 is a Brownian bridge; in the trend-stationary model,

Z 1
η2 ⇒ V (r)2 dr;
0

where
R1
V (r) = W (r) + (2r − 3r 2 )W (1) − (6r − 6r 2 ) 0 W (s) ds.

M.-Y. Chen I(0) vs I(1)

These tests suffer similar problems as the PP test. When ǫt
are highly correlated, the empirical sizes are too high if the
truncation lag n for s2T n is too small, say, 4. A large truncation
lag, however, has an adverse effect on the power of test.

M.-Y. Chen I(0) vs I(1)

Table 2: Asymptotic critical values of the KPSS tests.
Statistic 1% 2.5% 5% 10%
η1 0.739 0.574 0.463 0.347
η2 0.216 0.176 0.146 0.119

M.-Y. Chen I(0) vs I(1)

Panel Unit Root Tests

Consider the model

yit = ρi yit−1 + zit′ γ + uit , i = 1, . . . , N, t = 1, . . . , T, (4)

where zit is the deterministic component and uit is a

stationary process. zit could be zero, one, the fixed effects, µi ,
or fixed effects as well as a time trend, t.

M.-Y. Chen I(0) vs I(1)

Levin and Lin (1992), Levin, Lin, and Chu (2002)
Levin and Lin (LL) tests assume that ρi = ρ for all i and are
interesting in testing
H0 : ρ = 1 v.s. Ha : ρ < 1.
Denote ρ̂ as the OLS estimator of ρ in (4), Levin and
Lin (2002) show that
PN PT
√ √1 1
ỹi,t−1 ũit
N i=1 T t=1
N T (ρ̂ − 1) = 1 PN 1 PT 2
ỹi,t−1
N i=1 T 2 t=1
qP
N PT 2
(ρ̂ − 1) i=1 t=1 ỹi,t−1
tρ̂ = ,
se
N T
1 XX 2
s2e = ũ .
T N i=1 t=1 it
M.-Y. Chen I(0) vs I(1)
As T → ∞ and then N → ∞,
zit ρ̂
√
0 N T (ρ̂ − 1) ⇒ N(0, 2)
√
1 N T (ρ̂ − 1) ⇒ N(0, 2)
√ √
µi N T (ρ̂ − 1) + 3 N ⇒ N(0, 51/5)
√
(µi , t) N (T (ρ̂ − 1) + 7.5) ⇒ N(0, 2895/112)

M.-Y. Chen I(0) vs I(1)

As T → ∞ and then N → ∞,
zit tρ̂
0 tρ̂ ⇒ N(0, 1)
1 tρ̂ ⇒ N(0, 1)
√ √
µi 1.25tρ̂ + 1.875N ⇒ N(0, 1)
p √
(µi , t) 448/277(tρ̂ + 3.75N) ⇒ N(0, 1)
In case uit is stationary, the asymptotic distributions of ρ̂ and
tρ̂ need to be modified due to the presence of serial correlation.

M.-Y. Chen I(0) vs I(1)

Harris and Tzavalis (1999)’s Tests

Harris and Tzavalis (1999) also derived unit root tests for (4)
with zit = {0}, {µi }, or {µi , t} when the time dimension of
the panel, T , is fixed. This is typical case for micro panel
studies. The main results are
zit ρ̂ 
√ 
0 N(ρ̂ − 1) ⇒ N 0, T (T2−1)
√ 3


3(17T 2 −20T +17)

µi N ρ̂ − 1 + T +1 ⇒ N 0, 5(T −1)(T +1)3
√    2 −728T +1147)

(µi , t) N ρ̂ − 1 + 2(T15+2) ⇒ N 0, 15(193T 112(T +1)3 (T −2)

M.-Y. Chen I(0) vs I(1)

Im, Pesaran and Shin (2003)’s Tests
The tests of Levin and Lin (1992) are restrictive in the sense
that it requires ρ to be homogeneous across i. Im, Pesaran
and Shin (2003) allow for heterosgeneous coefficient of yit−1
and proposed an alternative testing procedure based on the
augmented DF tests when uit is serially correlated with
different serial correlation properties across cross-sectional
units, i.e., uit = pj=1
Pi
ψij uit−j + ǫit . Substituting this uit in
(4), we get
pi
X
yit = ρi yit−1 + ψij △yit−j + zit′ γ + ǫit , i = 1, . . . , N, t = 1, . . . , T,
j=1

M.-Y. Chen I(0) vs I(1)

The null hypothesis is

H0 : ρ i = 1

for all i against the alternative hypothesis

Ha : ρ i < 1

for at least one i. The t-statistic suggested by Im, Pesaran and

Shin (2003) is defined as
N
1 X
t̄ = tρ̂ , (6)
N i=1 i

where tρ̂i is the individual t-statistic of testing H0 : ρi = 1 in (5). It is

known that for a fixed N ,
R1
WiZ dWiZ
tρ̂i ⇒ h0 = tiT (7)
R 1 2 i1/2
0 W iZ

as T → ∞.
M.-Y. Chen I(0) vs I(1)
Im, Pesaran and Shin (2003) assume that tiT are i.i.d. and
have finite means and variances. Then
√  1 PN 
N N i=1 tiT − E[tiT |ρi = 1]
p ⇒ N(0, 1)
var[tiT |ρi = 1]

as N → ∞ by the Lindeberg-Levy central limit theorem.

M.-Y. Chen I(0) vs I(1)

Hence, the test statistic of Im, Pesaran and Shin (2003) has
the limiting distribution as
√
N(t̄ − E[tiT |ρi = 1])
tIPS = p ⇒ N(0, 1)
var[tiT |ρi = 1]

as T → ∞ followed by N → ∞ sequentially. The values of

E[tiT |ρi = 1] and var[tiT |ρi = 1] have been computed by Im,
Pesaran and Shin (2003) via simulations for different values of
T and pi ’s.

M.-Y. Chen I(0) vs I(1)

Combining p-value Tests
Let pi be the p-value of a unit root test for cross-section i, Maddala and
Wu (1999) and Choi (1999) proposed a Fisher type test as:
N
X
P = −2 ln pi (8)
i=1

which combines the p-value from unit root tests for each cross-section i
to test for unit root in panel data. P is distributed as χ2 with 2N
degrees of freedom as Ti → ∞ for all N . When pi closes to 0 (null
hypothesis is rejected), ln pi closes to −∞ so that large value P will be
found and then the null hypothesis of existing panel unit root will be
rejected. In contrast, when pi closes to 1 (null hypothesis is not
rejected), ln pi closes to 0 so that small value P will be found and then
the null hypothesis of existing panel unit root will not rejected.
M.-Y. Chen I(0) vs I(1)
Choi (1999) pointed out the advantages of the Fisher test: (1) the
cross-sectional dimension, N , can be either finite or infinite, (2) each
group can have different types of nonstochastic and stochastic
components, (3) the time series dimension, T can be different for each i
(imbalance panel data), and (4) the alternative hypothesis would allow
some groups to have unit roots while others may not. A main
disadvantage involved is that the p-value have to be derived by Monte
Carlo simulations.
When N is large, Choi (1999) also proposed a Z test,
PN
√1
N i=1 (−2 ln pi − 2)
Z= (9)
2
since E[−2 ln pi ] = 2 and var[−2 ln pi ] = 4. Assume pi ’s are i.i.d. and
use the Lindeberg-Levy central limit theorem to get

Z ⇒ N (0, 1)

as Ti → ∞ followed by N → ∞.
M.-Y. Chen I(0) vs I(1)
Hadri (1999)’s Test: KPSS Type
Let êit be the residuals from the regression:

yit = z′it γ + eit (10)

and σ̂e2 be the estimate of the error variance. Also, let Sit be
the partial sum process of the residuals,
t
X
Sit = êij .
j=1

Then the LM statistic is

1
PN 1
PT
N i=1 T 2 t=1 Sit2
LM = .
σ̂e2
M.-Y. Chen I(0) vs I(1)
It can be shown that
Z 
p
LM → E WiZ

R 2
as T → ∞ followed by N → ∞ provided E[ WiZ ] < ∞.
Also,
√ R 2
N(LM − E[ WiZ ])
q R ⇒ N(0, 1)
2
var[ WiZ ]

as T → ∞ followed by N → ∞.

M.-Y. Chen I(0) vs I(1)

Panel Unit Roots Tests in R: CADFtest

The asymptotic p-values of the Hansen’s (1995)

Covariate-Augmented Dickey Fuller (CADF) test for a unit
root are computed using the approach outlined in Costantini
et al. (2007).
1. CADFpvalues: p-values of the CADF test for unit roots
CADFpvalues(t0, rho2 = 0.5, type=c(”trend”, ”drift”, ”none”))

2. CADFtest: Hansen’s Covariate-Augmented Dickey Fuller (CADF)

test for unit roots

M.-Y. Chen I(0) vs I(1)

Panel Unit Roots Tests in R: plm

1. purtest: Unit root tests for panel data

purtest(object, data = NULL, index = NULL, test=
c(”levinlin”, ”ips”, ”madwu”, ”hadri”), exo = c(”none”,
”intercept”, ”trend”), lags = c(”SIC”, ”AIC”, ”Hall”),
pmax = 10, Hcons = TRUE, q = NULL, dfcor = FALSE,
fixedT = TRUE, ...)

M.-Y. Chen I(0) vs I(1)

M.-Y. Chen I(0) vs I(1)