Abstract
Panel data analysis aims to overcome the weaknesses of its alternatives: country-by-country analysis is usually based on short samples, there is a significant country-specific distortion in the data, and it leads to biased estimates, and the cross-section analysis neglects the time dimension. In last two decades, tests for non-stationary panels sparked a large body of literature both on tests theory and on various empirical studies in multiple areas of micro- and macroeconomic research. The most popular studies include topics such as growth, finance, exchange rates, fiscal matters, and international trade, but also popular are studies in tourism, energy, resource demand and supply, IT and technology spreading, politics, inflation, international trade and current accounts, stock markets, etc.
TopTests For Panel Non-Stationarity
Panel Unit Root Tests
Pioneering work in panel data on exploiting the information from cross-sectional dimensions in inferring non-stationarity is contributed to Quah (1994), who derives the normality of some cases of unit root regression in panels, demonstrating that coefficient estimators have a mixture of standard normal and Dickey-Fuller-Phillips asymptotics. Also found is that the standard normal distribution is a good approximation of both large N, small T, and large N, large T cases. This initial work is soon followed by a multitude of practical tests.
Key Terms in this Chapter
Non-Stationary Panel: A panel data set where some (or all) of the time series contain unit roots.
Panel Data: A data set with both cross-sectional and time series dimension. If all cross-sectional units are observed for the whole time period, the panel is balanced. Otherwise, if the time series for some of the cross-section units are not of the same length and there are missing values, the panel is unbalanced .
Panel Cointegration: Some (or all) of the non-stationary time series in the panel have a stable, long-run relationship.
Cross-Sectional Dependence: Time series for different cross-section units are correlated, as a result either from unobserved factors, or from spatial or spillover effects. Cross-sectional dependence is defined also as “weak” or “strong” (resulting from “weak” or “strong” common factors).