Suitable model for predicting a range-bound label?

wypee

Hi,

I am working on a problem where you need to predict percentage loss from a loan-default. I am wondering how it should be modeled:

1. Should no-default (0% loss) be predicted as a binary classification separately, given that it is the most frequent occurrence?
2. if I use linear regression to predict percentage loss, how can I make the label range-bound (0-100)?

Thanks in advance!

~Y