Originally Posted by TraderSeven

No special reason. Optimize on 2001 and test on 2003 is fine too.
The danger of optimizing is that the settings only work on the data you used.
If you optimize on year X and test on year Y you have proof that the settings 'always' work and not just on the backted data.

(Over) optimizing, also called curve fitting, can give you +100% on the data you optimized on and -100% on all other data.
Signs of over optimization are often that there are a lot of loosers and the profit comes from one big killer profit.