Custom Likelihoods in PyMC: One-Inflated Beta Regression for Loan Repayment

When a borrower takes out a personal loan, they might repay every penny, default entirely, or land anywhere in between. The interesting variable is the fraction eventually recovered: a number between 0 and 1 for each loan in the portfolio. Plot the distribution across thousands of loans and it looks like a smooth Beta curve with a tall spike bolted on at the right edge — a mass of borrowers who repaid in full. That spike is good news for the lender, but a headache for the modeller. Standard Beta