Statistics Glossary → Cross-sectional dependence (Pesaran CD)
Cross-sectional dependence (Pesaran CD)concept
In panel data, correlation between the residuals of different units (countries, firms, regions) in the same period; Pesaran's CD test checks whether such correlation is present.
Panel regression models usually assume that the error terms of different units are independent. Yet a global crisis, a shared policy change or spillovers between neighbours can hit many units in the same year, making the residuals correlated across units. Pesaran's CD test computes the correlation of the residuals over time for every pair of units and scales the average of these correlations into a statistic that is approximately standard normal under the null hypothesis of cross-sectional independence. It can be used even when the number of units exceeds the number of time periods.
If cross-sectional dependence is ignored, standard errors are usually too small and significance is overstated; if the common factor is correlated with the model's predictors, the coefficients can also be biased. When dependence is found, remedies include Driscoll-Kraay standard errors, time fixed effects or common-factor estimators. The CD test can miss dependence when positive and negative correlations cancel out.
An everyday example
Annual export growth in 30 countries is modelled with panel data. In 2009 the global financial crisis cut almost every country's exports at the same time. If the model does not include this common shock, the countries' residuals move together; the Pesaran CD test comes out significant and signals that the standard errors need to be adjusted for this dependence.
In MerQur
In Panel Data Regression, the 'Additional tests: Cross-sectional dependence (Pesaran CD)' option tests for cross-sectional dependence using the average correlation between the units' residuals.