Types of prior that might be asked to incorporate into the model:
- allowance for different datasets
- a lower bound below which stresses cannot go
- allowing for different dynamics at terms 1 and 2
- a belief that interest rate volatility will be higher/lower over coming year
- removal of drift eg long term downward drift seen prior to covid
- the yield curve will have a parallel downward shift of 30bps with probability 30%
parralell downward shift prior
Does a prior that says in 1 years time the yield curve will make precisely a parallel shift generate a zero likelihood... I think it possibly does ... i think the parameter values need to be able to express all of the history.
We therefore need priors that make soft (not hard exclusions).
the prior therefore needs to be continuous
normal, beta, log-normal distributions weight heavily weight preferred regions but don't exclude others