I am building a home lending model through Survival technique. Since, I wanted to use a lot of time varying covariates (change in marital status, cash balance, residence etc.), I decided not to go ahead with Linear/Logistic regression. My study period is of 2 years and there are 25 such points during study period when people actually took a home loan.

Now, I have a combination of dummy and continuous (age, assets etc.) variables predicting home loans. Proc PhReg in SAS provides parameter estimate (and their standard error, p-values) and Hazard ratio. Baseline statement also provides probability value for each combination of independent variable. My objective, however, is not only to study parameter estimates but also to create client list for targeting. Here are my 2 questions:

  1. Since my unique combinations of independent variables (of age, assets, marital status change etc.) are huge (literally endless on future unseen data), there is no point calculating all 25 probabilities for every combination. Is there a way I can calculate all 25 probabilities using parameter estimates so I can score only for few clients that I want

  2. I am thinking of coming up with a threshold probability. As soon as that is breached, I will say that this ID has significant chances of taking a home loan. So, 2 persons (each starting from survival probability = 1 at the beginning) can have threshold survival probability (let's say 0.25) after 50 and 100 days respectively based on their attributes that came significant in the model. So, I will prioritize person 1 for the next campaign. Is this approach fine or are there alternative approaches to rank order client for campaigns not so far away.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.