In this paper we compare the size and the power of four cointegration tests in heterogeneous panel data, with both varying intercepts and slopes. These tests are (i) Kao (1999) Tests (both Dickey-Fuller (DF) and augmented Dickey-Fuller (ADF) types of cointegration tests in panel data), (ii) McCoskey and Kao (1998) Test (Residual based LM Test for cointegration in panels), (iii) Pedroni (1997) Test, and (iv) Larsson, Lyhagen, and Lothgren (2001)?s likelihood-based (LR) panel test of cointegrating rank in heterogeneous panel models. Three of these tests, the Kao tests, the Pedroni test and the Larsson et. al. test, are constructed under the null hypothesis of no cointegration, while the other, the McCoskey and Kao LM test, is based on the null hypothesis of cointegration. Through Monte Carlo simulations, we examine whether the panel data procedures have more (or less) power as compared to single country time series data in the case of bi-directional changes in the frequency of observations. In this context, we study the consequences of increasing the frequency of the observations (e.g. from annual data to quarterly or monthly data etc) and the data spans (increasing time spans e.g. from 10 years to 20 or 30 years) as well as the effects of decreasing the frequency of the observations (e.g. from annual data to average of 3 years, 5 years or 10 years). To the best of our knowledge, no such study on the comparison of the power of panel cointegration tests when one changes the frequency of data has been done. Azhar Iqbal, Applied Economics Research Centre, University of Karachi, Karachi-75270, Pakistan. Phone: 092-21-9243168, Fax: 092-21-4829730, E-mail: drazhariqbal@yahoo.com
