Be careful... the more times you slice a dataset the more chances you give it to lie to you. :)
These kinds of effects are very unstable. You can actually make a better case for using shorter data histories than longer. For instance, repeat your work with 1990 - 1995, then 1995 - 2000...