search

ChrisWaller26 / bayesact

A combined frequency-severity model in BRMS which allows for left-censoring at deductibles.
4 stars 2 forks source link

Modify likelihood weighting #33

Closed ChrisWaller26 closed 3 years ago

ChrisWaller26 commented 3 years ago

Allow the user to modify the weighting of the likelihoods between frequency and severity.

Currently, the un-weighted default is used, which could cause bias towards optimising severity over frequency. There may also be other reasons a user may want to adjust the likelihood weighting, e.g. give less weight to smaller losses or less recent policy years.

ChrisWaller26 commented 3 years ago

I have discovered that the weights argument in BRMS already does this, so there is no need to add this functionality in ourselves

Atan1988 commented 3 years ago

yeah, it is a bit tricky to use the weights(wei) argument with non-linear specifications. Theoretically, it just treat it as additional observation, but I have a bit hard time wrap my head around how it works in real problems that I have tried to use it.