At work, I've produced a curve that will predict my net loss on a portfolio of accounts over 50 months. However the curve is not very sophisticated, mostly based on averages, no seasonality taken into account, etc. So it's not a perfect curve, but I'll work on that later.
What I'm focusing on right now, is how to predict the ending total net loss at month 50 after I start collecting data for the previous months. Since I'm only about 85% accurate with my curve today. What I'd like to do is at month 5 (when we start actually incurring net loss) I'd like to try to predict what the ending net loss will be at that time based on that new data.
What I started with was this. I know that by month X what percentage of my total net losses should have already occurred. Example, by month 12 I know that 60% of my losses are basically done. So If I take the total amount incurred already, say 1,000$, I can simply divide the two and come up with a prediction of what 100% will be (1000/.60=1666.67$)
The problem here is that this only becomes precise when I hit month 12. for months 5-11 the prediction is extremely volatile. Is there a different methodology that I could use reduce the volatility?
Advertise on Talk Stats