The econometric theory for panel data was largely developed for the case where N, the number of units (e.g. individuals or firms) was large, but T the number of time-periods was small. Time-series analysis dealt with cases where N was small but T was large. Recently, data-sets where both N and T are large and the same order of magnitude have become more common. Examples are the Penn World Tables, where the units are countries and financial data where the units are firms. These data allow much more flexible treatment of heterogeneity than small T panels but raise time-series issues such as unit roots and cointegration. This course will discuss the econometric theory for such panels, and consider the models and estimators available for stationary, integrated and cointegrated variables and examine the issues raised by between-unit dependence. The procedures will be illustrated with data for international financial variables.
There will be a detailed set of lecture notes and a set of practical exercises, which would include the estimation output from the practicals. The practicals will use the Stata software package and employ a number of macro panel datasets for analysis. The practicals will also involve some fairly straightforward programming of novel empirical methods.
Level of knowledge required: This course is designed for people who work with panel (longditudinal) data where there are a large number of both time-series and cross-section observations (large T, large N). It will involve applying a range of time-series methods (e.g. for unit roots and cointegration) in a panel context. People attending should have a good training in stastistics or econometrics including basic time-series and panel models.