Hello all,
Let's say for an insurance company, there are only two possible policy cancel reasons - Underwriting or Billing issues. I built two separate logistic reg to predict each cancel individually. In the UW cancel case, I removed the cancel due to Billing, which leaves only those that cancel due to UW and those never cancel; In the Billing cancel case, I removed the cancel due to UW, which leaves only those that cancel due to Billing and those never cancel. In an attempt to combine both attrition models into one, I plan to produce a combined probability of cancel, which is as follow: Prob of cancel = prob(cancel due to UW) + [1-prob(cancel due to UW)] * Prob(cancel due to Billing | Not cancel due to UW). Does this approach and/or formula make sense? Any other suggestions? Thanks in advance.