The paper develops a new test for panel unit root. The test suggested is a panel version of the Dicky-Fuller-type test. By taking full advantage of trending properties in data, the test is consistent at a rate faster than that considered in Levin, Lin and Chu (1997). The use of a pooled hyper-consistent estimator of unit root in panel regressions renders this feasible. The limit distribution of the test under the null, established by letting time series (T) and cross-sectional units (N) go to infinity is shown to be a standard normal. Our bootstrap tests are found to have correct rejection probability even for narrow and short panels, and to exhibit better power than the Im-Pesaran-Shin test statistics in large panels.
