How to handle covariates with different histories

As @junpenglao suggested in another thread, you can feed in the entire history, then have PyMC3 handle the missing values.

See https://github.com/pymc-devs/pymc3/blob/master/pymc3/examples/lasso_missing.py4