Relief pitcher stats are quite volatile for a couple of reasons -
- smaller sample size means luck/chance plays a much more significant role in the stats than for SP and position players
- they are often replaced with another reliever after giving up a few runs (as opposed to a SP that might give up 2 in the 1st inning, and another in the 2nd, and leave in the 7th after giving up a respectable 3). So they have less chance to 'modulate' their stats
Hence the year to year volatility mentioned on here
Need to use at least 2-3 years of data and rely more on scouting grades, pitch ratings
IMO