What returning production is really worth

Every August the same sentence runs on every preview show. This team returns most of its production, so watch out. It is true often enough to feel like a law and vague enough to mean nothing. How much is most of your production actually worth, in points, on the field? Our college football preseason rating puts a number on it, and the number is smaller and more honest than the takes make it sound.

What the input actually measures

Returning production in our model is a single field, percent_ppa, from our returning production table. It is not the share of starters back, and it is not the share of snaps back. It is the share of last season’s on-field value that returns, where value is measured in predicted points added rather than in bodies. A team that brings back a low usage backup counts that player for almost nothing. A team that brings back the quarterback who touched the ball on every snap counts him for a lot. The metric weights by what each returning player was worth, not by how many of them there are.

That distinction matters because the roster headlines lie. Returning 18 of 22 starters sounds like continuity, but if the four you lost were your quarterback, your top receiver, and both edge rushers, you did not return much value at all.

The weight

Inside the formula, returning production carries a coefficient of 5.8729. Read plainly, a full unit of returning production is worth about 5.9 rating points, where a rating point is a point of expected margin against an average opponent on a neutral field. Across the teams we rate for 2026, the input runs from about minus 0.57 at the bottom to 0.997 at the top, with the median team near 0.40. So the realistic spread from a gutted roster to a fully intact one is roughly one unit, and one unit is that 5.9 points.

Where does that rank? Third. Performance carryover, last season’s SP plus walked forward, is the biggest lever in the formula because it swings across tens of points from the top of the sport to the bottom. Talent, our four year blue chip ratio, carries the heaviest coefficient per unit. Returning production sits behind both. It is a real input, worth a real chunk of a win on most schedules, and it is not the headline the August shows make it.

What it looks like in 2026

The cleanest way to see the weight work is to find two teams the input disagrees about and watch the model follow it.

Iowa State enters 2026 with the better performance carryover of the two, a prior SP plus of 9.9 against San Diego State’s 6.7. On the carryover input alone, the Cyclones are ahead by about 1.8 rating points. But Iowa State returns almost nothing, a returning production figure of 0.017 after a heavy graduation and portal cycle, while San Diego State brings back 0.987, nearly all of its value. That gap is worth about 5.7 rating points in the model, all from this one input, and it more than erases the carryover edge. The rating ends up favoring San Diego State, 4.55 to 2.00. Same formula, and returning production is the reason the order flips.

Now the honest counterweight, because this input gets oversold in exactly one way. Returning production is not talent, and bringing everyone back from a bad team returns a bad team. Akron sits near the very top of the 2026 board in returning production at 0.918, and Akron still rates minus 7.96, a clearly negative number in the bottom quarter of the teams we rate. The model gives the Zips their 5-plus points for continuity and then the other inputs, a deeply negative carryover and almost no blue chip talent, drag them back down anyway. Continuity raises your floor by a few points. It does not change what you are.

Why the number is where it is

August overrates this input more than any other, and the fitted weight refuses to play along. The weight was not chosen to make a point. It fell out of ordinary least squares, regressing each team’s actual end of season SP plus on its preseason inputs across the 2021 through 2023 seasons, then held to the same value while the model was tested out of sample on 2024 and 2025. Returning production earned 5.87 points per unit because that is what returning production has been worth, on average, once you also account for how good the team was to begin with.

The limit is the same one that governs every number here. The whole rating misses actual win totals by about 1.9 wins per team on average, so no single input inside it, returning production included, is a lock. It is one honest term in a small honest formula. A team that returns everyone and rated well last year is a real preseason bet. A team that returns everyone and rated in the bottom quarter is Akron, and the model will tell you so.

The full field, every team we rate with its returning production, its rating, and its expected wins, lives at edgelabs.bet/cfb/power-ratings. Members get the win total leans first, with the sample and the error attached to each one. Founding access is open at edgelabs.bet/join.