How much historical data to use?

[tommedema]

Great library, I love the pseudocode and python implementation.

I read the paper at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3892335 but cannot figure out what the recommended number of historical days to use is. Is 3 enough, a month, a year? Is it possible to use too much data if I just want to estimate today's spread?