Hi,
We're trying to figure out a good way to move forward on refactoring date handling to hopefully improve / simplify the time series models' usage of dates in Statsmodels.
A little bit of background:
- Time series models currently support dates, primarily by leveraging Pandas functionality.
- The code was written quite a while ago, I think around Pandas v0.9 or so. As a result, it has a certain amount of code to work around issues in earlier versions of Pandas, and much of it is surely no longer required (this is one of the reasons for refactoring)
- The current works great for the primary use cases, but has an increasing number of corner cases since it's so old.
If you have any thoughts on what to do or what to avoid, I'm sure we'd find your input helpful.
Chad