OLS in statsmodels has currently no option to drop singular columns.
statsmodels OLS is using the Moore-Penrose generalized inverse, pinv, to solve the linear least squares problem. This means that the reported covariance has reduced rank. However, we can use estimable contrasts to get and test the effects for which the covariance is of full rank.
If a user wants to have a full rank solution with statsmodels, the user has to decide which of the collinear columns to drop.
One way to find collinear columns is described here http://stackoverflow.com/questions/13312498/how-to-find-degenerate-rows-columns-in-a-covariance-matrixhttps://stackoverflow.com/questions/13312498/how-to-find-degenerate-rows-columns-in-a-covariance-matrix
(Most likely an option to automatically drop collinear columns will be available in a future version of statsmodels.)
