variation of the ys. To solve these problems, Evans and Karras (EK, 1996)
suggest testing for the stationarity of the demeaned series, i.e.:
p
δyn - y)t = θn + Ψn(yn - y)t-1 + λni∆(Уп - y)t-i + νnt, (3)
i=1
where, φ = 0 if the economies diverge, and φ < 0 if they converge. EK
formulate a modified panel unit root test of eq. (3) that allows testing two
implications of Endogenous Growth Models (EGM), namely that φ = 0,
and θ 6= 0. They employ Monte Carlo simulation to provide approximate
distributions of τ(<^) and Φ(θ).
Evans and Karras (1996), however, dispense from two critical facts. Firstly,
they assume that νs are uncorrelated, an assumption that is likely to be in-
valid, especially for a finite cross-section of regional economies. Secondly,
they do not exploit the fact that φ can be equal to zero even if only some of
the economies diverge.
In this work, we intend to overcome some of these limitations exploiting
some of the recent advances in Panel Unit Root (PUR) tests. These tests
dramatically increase the power of univariate unit root tests by pooling cross
sectional time series data. One of the first PUR tests was developed by
Levin and Lin (1993) and then extended by Levin, Lin and Chu (LLC, 2002)
This test can be essentially seen as a pooled Dickey-Fuller test, or a pooled
Augmented Dickey-Fuller (ADF) test when lags of the dependent variable
are included to account for serial correlation in the errors.
p+1
δ yit = α*y it-1 + βt + X αj δ yit-j + ? it (4)
j=1
where yit = yit — yit. As in the univariate ADF, the null hypothesis is that the
series is non-stationary or integrated of order 1, I[1]. LLC derive a statistic
(t* ), which is distributed as a standard normal under the null hypothesis of
non-stationarity.
Although, the test accounts for individual effects, time effects and possibly
a time trend, it assumes that each cross-section in the panel shares the same