Projections: forecasting without a crystal ball

Every projection system — from fantasy sites to front offices — is elaboration on three moves you already know from this course. One: recent performance matters most, so weight the last three seasons (5/4/3). Two: everyone is partly luck, so regress toward the league average — the fewer PA a player has, the harder you pull him toward ordinary. Three: age matters — players improve toward ~29 and decline after.

Why 'boring' wins

That's the Marcel method (named, as a joke, after a monkey — the point being even a monkey applying these rules is hard to beat). StatLab runs its own Marcel: backtested on 2025, it beat the naive 'last season repeats' forecast by 18% on OPS error. What it can't do — by design — is see breakouts: its biggest 2025 misses were Cal Raleigh and George Springer, the seasons nobody's model saw.

The going-forward outlook on every player page is exactly this, and nothing more. Fancier systems add park factors, batted-ball data, and minor-league numbers — they gain a little accuracy, but the skeleton is always these three rules. Once you know that, projections stop being oracles and become what they really are: disciplined averages with the hype removed.