The Global Dimension to Fiscal Sustainability

where p(i) is the p-value associated with (ADF(i)) of the ADF test for the i cross
section, where ρi is the autoregressive parameter of the independent error processes
(see equation (9)). The test statistic examines whether H0: ρi = 1 ∀ i against HA: ρi < 1,
for some i.

The PANIC approach has a number of useful facets from the perspective of
our empirical study. For example, O’Connell (1998) suggests cross sectional
correlation causes standard pooled panel tests, such as that of Levin, Lin and Chu
(2002), to over reject the null hypothesis of a unit root. However, O’Connell’s GLS
data transformation requires that the common component is stationary.⁹ This may not
always be the case. The PANIC approach is advantageous since the common factors
and idiosyncratic components are consistent irrespective of whether they are
stationary or not: the unobserved components are estimated by first differencing the
data and then accumulating the estimates. Additionally, Jang and Shin (2005) provide
Monte Carlo evidence that Bai and Ng’s (2004) second generation panel unit root test
has preferable statistical properties to tests based on principle components such as
Moon and Perron (2004) and Phillips and Sul (2003). Due to the nature of subtracting
the factor in Bai and Ng (2004), there are more stable sizes under cross sectional
dependency and also OLS estimation (Jang and Shin, 2005).

3.2 Panel Estimation with a Stationary Factor

The Bai and Kao (2006) panel cointegrated regression estimator deals with
cross sectional correlation by utilising a factor model approach. In particular, this

⁹ Indeed SURE is infeasible in a situation in which the time dimension of the panel is less that the cross
sectional dimension (i.e. T < N). We consistently apply a factor approach.

12

<<   9   10   11   12   13   14   15   >>

More intriguing information