Suitable Model

[Cope_LA2]

Hi all,

I wonder if anyone could help me, i'm looking for an idea or two

I work in the financial sector and we are looking into ways of predicting the behaviour of customers following changes to rates on some of our products. The current process involves looking at previous rate changes and using knowledge about the current financial climate to come up with a figure, we're quite good at it actually, but we'd like to be more savvy and that where statistics comes in!

We currently use logistic regression in a similar situation but as this isn't really a lost/retained problem it doesn't seem suitable (the customer may choose to leave immediatly, after three months of after six and there's no real cut off date). We're interested if possible in predicting this loss pattern

I wondered if anyone had dealt with a similar problem and wether there was a statistical technique you'd reccomend?

All the very best
Laura

 

[Miner]

A proportional hazards model might be appropriate here.


http://en.wikipedia.org/wiki/Proportional_hazards_model

 

[Cope_LA2]

That's interesting Miner,

I've found a really good link for proportional hazard models here if anyone is interested

https://onlinecourses.science.psu.edu/stat507/node/81

I think it's probably going to be these i use though i've struggled in SPSS modeler to validate the models... Expected vs Actual graphs dont seem to work due to the censoring of customers. How would you go about measuring if one model was predicting well compared to another?