AComparative study of unit root tests with panel data and a new simple test（10）

G. S.Maddala and Shaowen Wu

(H. D.Thoreau: Walden)

1，Ineach experiment, we consider the cases

2，Table1 reports the simulation results of the model with constant termonly.

3，Overall,the performance of the LL test is the worst but it should be noted that asargued earlier, the power comparison between the LL test and the other twotests is not valid.

4，TheIPS and the Fisher tests are directlycomparable because they are tests of thesame hypothesis.

5，Whenreading Table 1, it should be kept in mind that the power comparisons are not appropriatebecause there are significant size distortions.

6，However,we have not computed size-adjusted power because some broad conclusions can be drawnfrom the figures presented, without looking at the size-adjusted power.

7，Themajor conclusion that one can draw is that for high values of T (50 or100) and values of N =50 and 100 the Fisher test dominates the IPS test in the sense that theFisher test has smaller size distortions and comparable power.

8，Thesame conclusions follow from Table 2, the case with trend.

9. Weshall see that this conclusion about the dominance of the Fisher test is reinforcedin the results presented in Table 3 later.

（学习待续）