As Lütkepohl  states his preface, "[b]y now [cointegration] related models and methods have become far more important for applied econometric work than, for example, vector autoregressive moving average (VARMA) models." In the context of cointegrated variables especially vector error correction models (VECM) are emphasized in .
While statsmodels supports VARMA models, it still lacks VECM (as suggested by one of the project's main contributors). The GSoC might be a good opportunity to introduce VECM-related functionality into statsmodels. This functionality includes the representation of a VECM, forecasting, estimation, and model diagnostics among other topics.
 Lütkepohl (2005) New Introduction to Multiple Time Series Analysis