The average duration is influenced by the length of a policy and the total number of policies. If you have 10 policies to start with in January 2008 the average duration will increase throughout the year. However a sudden influx of new policies will decrease sharply the duration, which creates a biased estimate of the average length for the policies that stick in the portfolio. Eventually, as the number of policies increases, the duration will tend to go down, which is not correct.

There was an idea of using median duration instead of average, which is somewhat better. Does anyone have a suggestion of what metric can be used, either an index or something else?

thanks, Valentin