Unit Root Tests
Unit Root Tests
Mei-Yuan Chen
Department of Finance
National Chung Hsing University
Φ(B) = 1 − φ1 B − φ2 B 2 − · · · − φp B p ,
Ψ(B) = 1 − ψ1 B − ψ2 B 2 − · · · − ψq B q ,
yt = α0 yt−1 + ut ,
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
and
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,
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).
Then, define the GLS detrended data, ytd using the estimates
associated with the ā:
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.
yt = xt at + zt′ b0 + ǫt , t = 1, · · · , T,
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 .
yt = a0 + b0 t + ǫt ,
yt = a0 + b0 t + yt∗,
Pt
where yt∗ = j=1 vj .
T −2 Tt=1 St2
P
ηi = , i = 1, 2,
s2T n
where
R1
V (r) = W (r) + (2r − 3r 2 )W (1) − (6r − 6r 2 ) 0 W (s) ds.
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)
H0 : ρ i = 1
Ha : ρ i < 1
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]
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:
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
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 → ∞.