On the Specificity and Performance of Panel Unit Root Tests
Stephanie de Silva
Here we examine the problems associated with cross-sectional dependency in panel unit root and stationarity tests. One method of accounting for cross-sectional dependency is an approximate linear factor model. The finite sample properties of three well-known methods using factors are compared and several new Information Criteria for the purpose of determining the number of common factors in a panel are suggested. Two new hypothesis tests are proposed by this thesis and both can account for cross-sectional dependency of a general form through use of a bootstrap. Both tests can allow for regime change under both null and alternative hypotheses. One test is based on the univariate Locally Best Invariant test for the null hypothesis of stationarity by Busetti and Harvey (2001). The other is for the null hypothesis of a panel unit root against a fractionally integratd alternative hypothesis. This allows the test to be applied to a broader class of models, some of which encompass asymptotically non-stationary processes. This test is based on the univariate test of Dolado, Gonzalo and Mayoral (2002). The tests and techniques examined by this thesis are applied to a panel of stock prices from the Australian prudential industry in order to examine the Market Efficiency Hypothesis.
Andrew Tremayne and Vasilis Sarafidis.