March 14, 2025. The college football transfer portal’s spring window is open for fourteen days, and within those two weeks, 2,847 FBS players will enter their names — more than the entire FCS plus Division II annual movement combined, condensed into the fortnight after spring practices end. By March 28, the same window will close, and roughly 71% of those players will have committed to a new program. Some of those moves will be lateral. Some will be ascensions: a Group of Five All-Conference defensive end going to the SEC. Some will be drops: a struggling Power Four backup quarterback finding a starting job at the FCS level. By Memorial Day weekend, the rosters that will line up on opening Saturday in late August have effectively been redrawn, and the public analytical conversation about each team’s 2025 season is starting over from scratch. The math that tries to model this — what a portal move is worth, how to value an inbound transfer, how to project the displaced production at the player’s old school — is, in 2026, the single most underdeveloped corner of college football analytics.
The transfer portal era has, in a real sense, broken the analytical framework that worked for college football from 2010 to 2018. Returning production, the strongest single predictive variable in CFB, was originally built on a stable assumption: that the major year-to-year roster flows were graduation, NFL Draft, and small numbers of position-specific transfers. The portal, particularly in the post-NIL era, has turned that assumption on its head. A team can lose its starting quarterback in December, replace him with a portal addition in January, and have a wholesale offensive identity by April. The standard projection models, which were built on snap-by-snap continuity of returning players, do not handle this gracefully. The newer models that have tried to incorporate portal flows have produced rankings that are interesting, sometimes prescient, and frequently wrong. The math, in short, is being built in public, in real time, by analysts working with incomplete data against a moving target.
I have been writing about college football analytics since 2018, which means I have spent my entire career in the portal era, and the modeling challenge that has occupied the most quiet conversations in the analytics community is the one this article is going to unpack. Transfer portal math — how the public models try to value player moves, where the math works, where it fails, and what the next generation of CFB analytics needs to figure out, is the subject of this article.
The origin: where transfer portal math came from
The NCAA transfer portal was created in October 2018, replacing the old “permission to contact” system with a more transparent database where any player could enter to signal intent to transfer. The original rules retained the requirement that most transfers sit out a year of competition unless they qualified for a specific waiver. The 2021 NCAA reforms — sparked by the NIL settlement and a broader push for player movement rights — eliminated the sit-out requirement for first-time transfers. By 2022, the portal was effectively a free-agency mechanism, and the volume of moves exploded.
The analytics community caught up unevenly. The first serious effort to value portal moves came from Phil Steele’s magazine in 2019, which tracked notable transfers in a portal addendum but did not attempt to quantify their predicted impact. Pro Football Focus began publishing portal rankings around the same time, applying its existing player-grading framework to incoming and outgoing players. 247Sports launched its transfer-portal index, which used recruiting-rating math to assign each transfer a “value” similar to its recruiting class composite.
By 2023, the better models — SP+ chief among them — had started incorporating portal flows directly into the returning-production calculation. The math, in its simplest form, treated an inbound transfer’s prior-season production as a quality-adjusted addition to the new team’s returning production, while treating the outbound transfer’s production as a subtraction from the old team’s. The quality adjustment was, and remains, the methodological battleground.
The unsolved problem at the heart of transfer-portal math, in 2026, is how to value cross-conference moves. A player whose production was earned at a Group of Five program does not, in general, replicate that production at the Power Four level. The adjustment factor — how much to discount Group of Five output when projecting Power Four contribution — has been the subject of ongoing debate among analytical writers, and the most defensible answer (“it depends, by position and by program”) has not been satisfying to anyone who wanted a clean number.
How transfer portal math works: in plain language
The structural framework, applied by the leading public models, has three components.
First, prior-season production. The model identifies the player’s snap count, contribution metrics (passing yards, rushing yards, receiving yards, defensive tackles, sacks, etc.), and quality-adjusted output from his most recent season at his old program. The data is the same data that powers returning-production calculations for the player’s current team.
Second, league/conference adjustment. The model applies a quality multiplier to the prior-season production based on the strength of the player’s prior conference relative to his new conference. A wide receiver who caught 60 passes for 850 yards in the Mountain West, transferring to the SEC, has his projected SEC contribution discounted by some factor. The factor varies by position and by the specific quality differential between conferences. The current state of the art for these adjustments is, in my opinion, roughly 60-70% of the prior production translating for skill positions, with linemen and defensive backs at slightly higher rates.
Third, fit and scheme considerations. The harder layer, mostly unmodeled in public analytics. A quarterback who excelled in a quick-game offense may not produce the same way in a vertical-passing system. A defensive end who recorded 11 sacks in a 4-3 scheme may have a different role and different output in a 3-4. These factors are recognized by analytical writers but rarely quantified in the public models. The proprietary work inside clubs and at major programs does include scheme adjustments; the public versions mostly do not.
The critical component: the cross-conference adjustment
The single most consequential modeling choice in transfer-portal math is the cross-conference adjustment. The decision determines, in practice, how much credit a Power Four program gets for adding a Group of Five star versus a peer Power Four contributor, and how much penalty a Group of Five program suffers when one of its stars leaves for a more competitive league.
The empirical work on this question — published primarily by Connelly and a handful of analyst-bloggers — suggests the adjustment factors look roughly like this, from my reading of the data through the 2024 season:
An SEC, Big Ten, or top-quartile ACC player moving laterally within the Power Four conferences retains roughly 75-85% of his prior production, depending on position. Quarterbacks retain more (scheme being the bigger variable). Defensive backs and linebackers retain less (athletic context shifts more).
A Group of Five All-Conference player moving to a Power Four program retains roughly 50-65% of his prior production. The discount is largest for positions where the speed-of-game differential is highest (skill positions on offense and defense). It is smallest for offensive linemen and special teams players.
An FCS player moving to FBS retains roughly 30-45% of his prior production. The discount here is so large that the move is almost always best modeled as a depth-chart addition rather than a meaningful production boost.
Transfer portal math vs the alternatives: a comparison
The major public approaches to valuing transfer-portal additions, side by side:
| Approach | What it tries to do | Strengths | Weaknesses |
|---|---|---|---|
| 247Sports composite | Star-rating adjusted by transfer year | Simple, widely cited | Treats portal additions like high-school recruits; ignores game data |
| PFF transfer grades | Position-specific grade carried forward | Detailed individual evaluation | Subjective scoring, proprietary methodology |
| SP+ returning production net | Quality-adjusted production net flow | Integrated into broader projection model | Less granular at the individual-player level |
| On3 NIL Valuation | Market-based price for player services | Reflects current real-world demand | Volatile, partly speculative, not directly tied to on-field production |
| Independent analyst models | Custom scoring, often by beat reporters | Position-specific, context-aware | Inconsistent methodology across writers |
The honest version of portal coverage triangulates across two or three of these. A player with a high 247 composite, decent PFF grades, and reasonable on-field production at his prior program is a coherent positive signal. A player who scores high in one framework and low in another deserves a longer conversation about why.
What the data needs: inputs
Transfer-portal valuation requires several layers of data. The minimum inputs are play-by-play production data for every FBS and FCS team, roster tracking with daily portal updates, and position-specific snap counts for every player at the prior program.
The data is, in 2026, more available than it was five years ago but still fragmented across multiple sources. 247Sports maintains the most-cited portal database. On3 runs a competing tracker. The cfbfastR package provides play-by-play data for FBS teams. PFF provides position-graded player evaluations. The NCAA’s transfer portal database itself is the source of truth for entry and commitment events, though its API is limited.
For the cross-conference adjustments specifically, the necessary input is a leaguewide quality differential measurement that the analyst can apply consistently. SP+’s conference SP+ averages, published annually, are the most-defensible public version. Other models use their own quality measurements (FPI conference rating, Massey conference rating, etc.) which produce somewhat different adjustment factors.
The hardest input to obtain — and the one most aspirational portal models lack — is position-specific scheme tagging. Whether a quarterback’s prior offense was a quick-game spread or a vertical pro-style is, in theory, identifiable from film charting; in practice, no comprehensive public database tags this. The proprietary versions inside clubs and major programs exist; the public-facing analytical writing in 2026 still mostly works without this layer.
Building the analysis: a working framework
The practical workflow for evaluating a team’s portal class:
- List the inbound and outbound transfers with their prior-season production and prior program.
- Apply the cross-conference adjustment using the published quality factors (75-85% for lateral Power Four, 50-65% for Group of Five to Power Four, etc.).
- Calculate the net production change: inbound (adjusted) minus outbound (full value if going to comparable program).
- Convert to SP+ points. The rough rule of thumb: every 5% of net returning-production gain translates to roughly 1-2 SP+ points of projected improvement. The conversion is noisy but defensible.
- Apply scheme and fit adjustments, qualitatively, based on coaching context. A team adding three skill players to an offense that runs a system they fit translates to more impact than the raw production math suggests; a poor fit subtracts from the projection.
- Cross-reference with the model’s automated output. SP+’s returning-production rankings, published in May, incorporate most of this work. The personal calculation is mostly a check on the model.
Where this gets weird: common mistakes
The traps in portal analysis.
The “instant savior” trap. Mainstream coverage of a high-profile portal quarterback addition often treats him as a guaranteed upgrade, regardless of fit. The data is messier than that. Even top-rated portal QBs underperform their projections in roughly 30% of cases, due to scheme mismatch, injury, or simple chemistry issues with new receivers and offensive linemen.
Ignoring outbound losses. The same coverage that celebrates inbound additions often glosses over outbound departures from the same program. A team that signed a top-25 portal QB but lost its leading receiver and starting left tackle in the same window may not, on net, be improved at all.
Treating Group of Five production as Power Four equivalent. The cross-conference adjustment is large for skill positions. A wide receiver who put up 1,200 yards in the Mountain West will, in expectation, post 700-800 in the SEC. The careful writer applies the discount; the lazier coverage cites the Group of Five number as if it were SEC-equivalent.
Single-season production myopia. A portal transfer’s prior season is one data point. A player coming off a career-best year is, statistically, more likely to regress toward his career mean than to repeat. Career-spanning production is the more reliable input. Most public portal coverage uses only the most recent season.
NIL valuation as production proxy. On3’s NIL valuations have crept into portal coverage as if they were a measurement of on-field expected production. They are not. NIL valuations are market prices that reflect a mix of on-field expected production, marketing potential, locker-room presence, and bidding-war dynamics. A high NIL valuation is suggestive but not directly comparable to projected SP+ contribution.
When transfer portal math shines: use cases
The applications where the math has earned its keep:
Preseason projections for portal-heavy programs. Teams like Ole Miss, Colorado, Florida State, and Michigan have, in various seasons, built rosters that were 30-40% portal-sourced. The traditional returning-production calculation, ignoring portal flows, would have projected those teams as rebuilds. The net-flow version, with cross-conference adjustments, has produced much more accurate projections.
Bowl-game opt-out evaluation. By December, the major bowl matchups feature rosters that already differ meaningfully from the rosters that played the conference championship games. Portal entries, NFL Draft declarations, and opt-outs combine to leave many bowl-game starting lineups with 4-6 missing contributors. The math of net production at the line of kickoff is, increasingly, what the betting markets are pricing.
Mid-spring program assessments. By April, the spring portal window has closed and rosters are mostly settled for the upcoming season. The net portal flow combined with returning production produces a much cleaner read on each program’s expected quality than the late-season narrative coverage from the previous fall.
Identifying overlooked programs. A Group of Five program that signed two underrated portal additions while losing only depth-chart pieces is, by net production, likely to be better than its preseason poll ranking suggests. The mathematics-first coverage frequently identifies these programs three to four months before the polls do.
A working example: the 2024 Indiana Hoosiers
Indiana’s 2024 season is the most-cited transfer-portal case study of recent vintage. Curt Cignetti took over as head coach in November 2023, inherited a roster that had finished 3-9, and proceeded to add 24 portal transfers over the offseason, including six players who had been All-Conference selections at their prior programs. The transfer class included a quarterback (Kurtis Rourke from Ohio University), receivers, an offensive line, and a complete defensive overhaul. The traditional returning-production calculation for Indiana entering 2024 was, on the surface, dismal. SP+’s preseason projection nevertheless rated them as a top-50 team, partly on the strength of Cignetti’s recent track record at James Madison and partly because the net-flow portal calculation suggested a meaningful upgrade.
The Hoosiers went 11-1 in the regular season, beat ranked opponents, and finished the year as one of the most efficient teams in the country by SP+. The portal math had captured something the human consensus had missed: a small-school coach with a coherent scheme, combined with 24 carefully-chosen portal additions, can produce a team that resembles a multi-year build rather than a desperation roster patch. The model was not perfect — Indiana’s playoff loss to Notre Dame exposed depth-chart issues the projection had not fully anticipated — but the gap between the preseason media narrative (“Indiana will be terrible again”) and the in-season reality (one of the surprise teams of the year) was a portal-math gap.
What 2024 Indiana also illustrated, and what is the harder lesson for the modeling work, is that portal classes are coherent objects, not collections of individual moves. The same 24 players, brought into a poorly-organized program, would not have produced the same result. The math of the individual moves was reasonable; the multiplier effect of the class as a unit is the part the models are still learning to capture.
The limits: what transfer portal math cannot tell you
The honest version of portal writing names the limits.
Portal math cannot tell you whether a transfer will fit a new scheme. The cross-conference adjustment captures part of the translation; the scheme-specific adjustment is mostly unmodeled in public analytics. A quarterback who excelled in a quick-game offense may struggle in a vertical-passing scheme regardless of conference level. The writer who incorporates scheme context in addition to the math produces more reliable analysis than the pure number alone.
Portal math cannot model locker-room chemistry. A team that adds 20 portal transfers in a single offseason has to integrate those players into an existing culture, manage NIL disparities between transferred-in and homegrown players, and rebuild defensive communication and offensive timing. The on-field impact of these soft factors is real and the models cannot capture them.
Portal math is, fundamentally, working with a small and unstable historical dataset. The post-NIL portal era is only three or four full cycles old. The published cross-conference adjustment factors are calibrated on a sample of moves that may not be representative of the equilibrium the sport is heading toward. Updates to the adjustment factors are likely as more data accumulates.
Portal math cannot account for ongoing rule changes. The NCAA continues to revise transfer rules, NIL regulations, and eligibility standards on roughly annual cycles. A model calibrated to the 2023 portal environment is operating with different assumptions than the 2026 environment. The adjustment factors are themselves a moving target.
One additional limit: the public-facing portal analytics ecosystem remains less mature than the equivalent for NFL or NBA work. The data is harder to access, the methodologies less standardized, and the writing layer thinner. Expect this to improve over the next several seasons as the demand for portal coverage continues to grow.
Frequently asked questions
How big is a typical transfer portal class for a Power Four program?
In the 2025 cycle, Power Four programs averaged roughly 15-20 inbound transfers and 12-18 outbound, depending on coaching stability and roster needs. Teams in coaching transitions can have 30+ moves in a single offseason. Group of Five programs typically have smaller absolute numbers but larger ratios — a 65-scholarship roster turning over 20 spots is more disruptive than a 95-scholarship Power Four program doing the same.
Does a high 247 transfer rating predict on-field production?
Loosely. The 247 transfer composite correlates with subsequent on-field production at roughly r = 0.35, which is meaningful but not predictive in the same league as returning production (r = 0.55). The composite weights recruiting pedigree and prior-program quality heavily; on-field production from the player’s most recent season is partially incorporated but not dominant. Cross-checking with PFF grades and snap-weighted production produces a more reliable signal.
Why is the cross-conference adjustment so large for skill positions?
Speed of game and athletic differential. A wide receiver running a 4.48 forty-yard dash who dominated Conference USA defenses is, against SEC defenders running 4.40 themselves, a different player. The yards-after-catch component of his prior production was substantially driven by athletic separation that no longer exists at the higher level. Offensive linemen and special teams players see smaller adjustments because the athletic differential matters less for their primary tasks.
Where can I see public portal data?
247Sports’ transfer-portal database is the most-cited public source. On3 maintains a competing tracker with their own NIL valuations layered in. PFF provides position-specific grading for portal moves (subscription required). SP+’s annual returning-production rankings, published at Connelly’s Study Hall newsletter, incorporate net portal flows for each FBS program.
Sources and further reading
- 247Sports transfer portal database — the most-cited public tracker for inbound and outbound moves, with composite ratings.
- On3 transfer portal tracker — competing public database with NIL valuations integrated.
- Bill Connelly’s Study Hall — the analytical writing that integrates portal flows into preseason projections.
- Pro Football Focus — position-specific player grading that informs portal evaluation (subscription).
- cfbfastR R package — the open-source data layer for play-by-play that powers most public CFB analytics work.
The 2,847-player portal cycle that opened this article is no longer an exception; it is the new normal in college football. The math that tries to make sense of it is, in 2026, still being built. The writers who do this work carefully — Connelly, the PFF charting team, a small group of beat reporters who have learned to combine traditional reporting with the data — are producing analysis that survives the offseason in ways the take-of-the-day coverage does not. For the broader frame on how returning production fits into the picture, our SP+ explainer is the natural companion piece. The portal era is the project. The math is still catching up.
