Your 22 may be a fast trainer, and your 18 may be a slow trainer, so despite difference in age (and most notably difference in age buckets, i.e. 18 - 21 vs 22-23) they have similar training efficiency.

When buying players, I focus on 18-19 yo with > 80%. Upon purchase, I subject a player to training, typically around 150% condition loss, and divide the % skills by condition loss, which yields player's Training Efficiency Factor (TEF) - this has been discussed at length here. My numbers vary a bit compared to AI but it's also important to take into account training level.

At my current level I find TEF for 5* 18-21 year old above 25% as good (after level 3 training drills).

I do reckon age is a very significant factor, however have not paid that much attention to market value - perhaps I should ...?