At least my initial analysis of the data seems to say that degrading instantly is better than any other time constant đ. (May just be some nasty bug in the way weâre calculating probabilities?)
Discussion
Err, no, sorry, thatâs wrong. Degrading instantly is bad (the learning does help!), but the model itself is worse than the naive âI dunno, 50/50 alwaysâ âmodelâ, even when you learn.
(The LDK âhistorical modelâ, however, seems to do okay, keeping histograms of the liquidity bounds)