Following the Talent: Player Mobility and Talent Concentration in College Football

Abstract

We examine player mobility and talent concentration in college football using weekly performance grades, rosters, recruiting records and draft outcomes from 2015–2026. Comparing 2016–2019 with 2022–2025, the annual share of returning players’ proven value that changed programs rose from 1.9% to 23.5%. Of transferred value, 77% went to a higher-ranked program (95% CI 75–80%). The top-50 share of measured FBS talent, including recruit-implied contributions, remained similar at 42.9% and 43.1%. Crediting transfers to their origin programs reduces the post-period share by 1.7 percentage points. Net transfer value is associated with changes in team strength; G5-to-Power movers have higher subsequent draft rates than observably similar stayers. A career-stage comparison suggests that mobility growth extends beyond players affected by the 2021 eligibility reform, but does not separately identify NIL’s effect. These exploratory results describe greater reallocation of developed talent within a broadly persistent hierarchy. Overlapping reforms, selection into transfers and limited season-level samples constrain causal interpretation.

Keywords: college football; name, image and likeness; transfer portal; competitive balance; talent allocation; player mobility

JEL: Z22, J61, J62, L83

1. Introduction

For most of its history, college football restricted the movement of its labor. An undergraduate who changed FBS schools lost a season of competition, and players could not be paid beyond the cost of attendance. Both restrictions fell in 2021. In April the NCAA Division I Council extended the one-time transfer exception to football, so a first-time transfer could play immediately (NCAA, 2021). On July 1, following the Supreme Court's decision in NCAA v. Alston (2021) and a wave of state laws, the NCAA allowed athletes to earn money from their name, image and likeness. Later changes widened both channels: unlimited immediate-eligibility transfers from 2024, and direct revenue sharing under the House v. NCAA settlement from the 2025–26 academic year.

Public debate about these changes has focused on two predictions: that money would let wealthy programs hoard talent, and that free movement would hollow out smaller programs. Both are claims about the allocation of talent, and both are hard to test with the data usually available. Recruiting rankings describe players at entry, before most of their ability is revealed. Transfer counts weight a walk-on the same as an all-conference starter.

This paper measures the allocation of proven ability directly. We use weekly performance grades for FBS player-seasons from 2015 to 2026, which let us value each player by what he did on the field the season before he moved. We link these grades to rosters, recruiting ratings, the NFL draft and the transfer portal, and we place every program on a single Division I ranking that spans FBS and FCS. This lets us ask four questions. How much proven talent moves, and in which direction? How does mobility growth differ across career stages affected differently by the 2021 eligibility rule? Has talent become more concentrated at the top? And what does the movement do to teams and to players?

The contribution is a performance-weighted accounting of player mobility alongside measures of talent concentration, team strength and draft outcomes. The career-stage comparison offers an exploratory decomposition of mobility growth. The analysis describes associations rather than separately identifying the effects of NIL and transfer reform.

2. Institutional background

Table 1 lists the rule changes relevant to our sample. Three features matter for identification. First, graduate students had long been able to transfer and play immediately, so the 2021 eligibility change did not change their immediate-eligibility entitlement. Second, the transfer and NIL changes arrived within three months of each other in 2021, so a simple before-and-after comparison cannot separate them. Third, the NCAA granted an extra season of eligibility to athletes who competed in 2020, which enlarged the pool of fifth- and sixth-year players from 2021 to about 2024.

We define three periods. The pre-period is the 2016–2019 seasons. The transition years 2020 and 2021 are shown but excluded from comparisons, because 2020 was shortened by the pandemic and 2021 straddles the rule changes. The post-period is 2022–2025. The 2026 season is reported where available; it covers roughly the first seven weeks.

3. Related literature

The economics of sports labor markets begins with Rottenberg's (1956) invariance principle: if players can move to the clubs that value them most, the distribution of playing talent depends on those valuations, not on who holds the rights to players, so freeing the market changes who collects the rents more than it changes competitive balance. Fort and Quirk (1995) formalize the conditions under which the principle holds and survey the evidence from professional leagues; Szymanski (2003) reviews the contest-design literature that followed. Rosen (1981) explains why small differences in ability can command large differences in pay when the best performers can serve a large market, which is relevant to a sport where a few programs capture most broadcast and playoff revenue.

Work on NIL itself is recent. Li and Derdenger (2025) study recruiting before and after NIL and find that talent at entry became more dispersed: lower-ranked programs signed more highly rated recruits, and betting spreads narrowed after controlling for talent. Owens, Rennhoff and Roach (2025) find that a school's NIL activity predicts whether recruits accept its offers, and conclude that NIL helps elite programs but is unlikely to let others close the gap on its own. Both studies measure talent at the point of signing.

Research on transfers is mostly descriptive. Dohrn and Lopez (2022) follow quarterbacks who transferred before the 2019 season. Pifer (2026) applies network analysis to FBS transfers from 2019 to 2024 and finds that non-autonomous conferences gain transfers by raw count, while autonomous programs gain once transfers are weighted by talent and experience. Our results agree with that reversal and extend it in four directions: we value players by their measured performance rather than their recruiting rating, we include FCS and the pre-portal years, we compare mobility growth across career stages, and we connect flows to team ratings and draft outcomes.

4. Data and measurement

4.1 Sources

Performance grades. Pro Football Focus (PFF) grades every player on every play of every graded game. We use the weekly offensive and defensive summaries, which give each player's snap counts and game grade on a 0–100 scale. Our sample has 117,422 FBS player-seasons from 2015 to 2026. PFF also identifies each player's team by a franchise identifier, which we map to divisions and conference membership by season.

Rosters, recruiting, portal and draft. From the CollegeFootballData.com API we use season rosters (one stable athlete identifier per player), recruiting ratings and star counts, transfer portal entries for 2021–2026 (18,896 records), and every NFL draft selection from 2015 to 2026 with the player's college identifier.

Team strength. We use two independent team ratings. SP+ (Bill Connelly) is a widely used efficiency rating for FBS teams. The Athletic Vector Rankings (AVR) Division I board rates FBS and FCS programs on one scale using opponent-adjusted game and drive results; we use the final board of each season, 2015–2025. AVR is the author's own system (see the conflict-of-interest statement).

4.2 Linking players across sources

PFF and CFBD use different player identifiers. We match them within season and team on normalized names (accents, punctuation and suffixes removed), falling back to last name plus first initial when that is unique on both sides, and assign each PFF identifier the roster identifier it matches most often across seasons. Between 78% and 91% of PFF player-seasons are matched in every season from 2015 to 2026 (Appendix Table A1), and 85.4% of draft picks link to a graded player. Transfers in the performance data do not depend on this link, because PFF's own identifier follows a player across schools.

4.3 Definitions

Season grade. A player's season grade is the snap-weighted mean of his weekly game grades on his main unit (offense or defense, whichever he played more), and his team is the one where he played most of those snaps. A regular played at least 300 snaps on that unit. A star is a regular whose season grade is at least 69.5, the 90th percentile of FBS regulars in 2016–19, held fixed for all years.

Proven value. We summarize a player's demonstrated contribution with

where g is the season grade and n the snap count. We report value in starter-equivalents (SE): one SE is 10.5 units, the value of a 75-grade player who played 700 snaps. Value is always measured in the season before a move, so it reflects information available when the move was made. The threshold of 60 and the linear form are choices; Section 7 shows the main results under five alternatives.

Transfers and rank bands. A transfer is a player who appears for a different program than in the previous season. Power programs are members of the SEC, Big Ten, Big 12, ACC and (through 2023) Pac-12, plus Notre Dame; the rest of FBS is the Group of Five (G5). Conference realignment is not a transfer, because the player's program does not change. We also place each program by its rank on the previous season's final AVR Division I board, in bands of 1–10, 11–25, 26–50, 51–100, 101–150 and 151 or lower (mostly FCS).

Full talent including recruits. Performance grades do not value players who have not yet played. For Section 6.4 we assign every rostered FBS player an expected contribution for the current season. Players with at least 150 snaps the previous season are valued by the mean current-season value of players in the same decile of prior value; everyone else is valued by the mean current-season value of unproven players with the same star rating and years since signing. The mapping is estimated once on 2016–2025 and applied to all seasons. Unproven players account for 28–31% of the resulting talent stock.

4.4 Coverage

FBS coverage is complete throughout: 127–138 teams and about 3,500–3,800 regulars per season, except 2020 (2,460 regulars in shortened seasons) and 2026 (partial). FCS coverage is not. PFF graded few FCS games before 2019, with fewer than 80 FCS regulars a season in 2015–2017 against more than 3,000 from 2021. We therefore report FCS flows only from 2021, and exclude FCS-ranked programs from comparisons that span both periods.

5. Empirical strategy

5.1 Before-and-after comparisons

Our descriptive results compare season-level means between the pre- and post-periods. With four seasons on each side, an exact two-sided permutation test over seasons has a minimum attainable p-value of about 0.029. For estimates built from player or team records we report 95% cluster-bootstrap intervals (Cameron, Gelbach & Miller, 2008); clustering units are specified in the figure and table captions. Program-clustered intervals allow for dependence among a program’s players and seasons. The analysis was not pre-registered. We report every outcome we examined in the main text or the appendix, and we flag results whose intervals are wide or whose p-values are near conventional thresholds (Gelman & Loken, 2014).

5.2 Comparing mobility across career stages

The 2021 rule removed the sit-out cost for undergraduates but did not change the rules for graduates, who could already play immediately. Other forces that arrived in the same window (NIL income, the culture of the portal, the extra eligibility year, coaches' recruiting from the portal) plausibly affect both groups. Let m_k,t be the share of players in group k who change programs in period t, for undergraduates after their first or second season (U) and players who have completed four or more seasons since signing (G, mostly graduates). We estimate the rate ratios RR_k = m_k,post / m_k,pre and

θ is a difference-in-differences on the log scale, and κ is the share of the undergraduate increase that is matched by the graduate increase. We use the multiplicative form because pre-period undergraduate rates are close to zero, where an additive comparison is dominated by the group with the higher baseline; we also report the additive estimate. The identifying assumption is that, absent the rule, undergraduate and graduate mobility would have grown by the same proportion. That assumption is strong. Graduates may respond more to money because they are closer to the end of their careers, and the extra eligibility year raised their numbers. We therefore read κ as an approximate decomposition rather than a causal estimate of the effect of NIL. The pre-period for this design is 2016–2018, because the 2019 roster data contain too few fifth-year players to estimate a graduate rate reliably (Appendix A2).

5.3 Team outcomes

For program t in season s, let R be its AVR rating and Net the proven value of players it gained by transfer minus the value it lost, in SE. We estimate

The lagged rating absorbs regression to the mean. Because Net is built only from the previous season's performance, it uses no information from the outcome season. It is not randomly assigned: programs with new coaches or rising budgets may both attract transfers and improve for other reasons. We interpret γ as a predictive association. In Section 6.7 we decompose incoming value by its source and type.

5.4 Player outcomes

To compare movers with similar players who stayed, we use coarsened exact matching (Iacus, King & Porro, 2012). Each mover is compared with stayers from the same origin category in the same stratum of prior grade (bins at 55, 60, 65, 70, 75 and 80) and, for draft outcomes, the same unit. The estimate is the mean over movers of the difference between the mover's outcome and the mean outcome of stayers in his stratum,

with bootstrap confidence intervals (the clustering unit is specified in each figure caption). Outcomes are the change in season grade (among players who remain regulars) and selection in a subsequent NFL draft. Movers are selected on characteristics we do not observe, such as a strong spring or an interested Power coach, so these estimates describe how movers fare relative to observably similar stayers, not the effect of moving on a randomly chosen player.

6. Results

6.1 The volume of movement

FBS-to-FBS transfers rose from an average of 219 a season in 2016–19 to 1,642 in 2022–25 (permutation p = 0.028), and to 2,423 in 2026. As a share of FBS roster spots, transfers rose from 1.5% to 10.4%. The share of each roster that returned from the previous season fell from 63.5% to 55.2% (program bootstrap CI for the change: −9.8 to −7.2 points).

Counts understate the change. Figure 1 weights each mover by proven value. In 2016–19, 1.9% of the proven value of returning players changed teams each year. In 2022–25 the figure was 23.5%, rising every season to 35.3% in 2025 and 38.5% so far in 2026. Movers are now disproportionately the players who had already played well.

6.2 The direction of movement

Talent moves up the ladder. Weighted by value, 77% of transfers in 2022–25 went to a program ranked higher on the previous season's AVR board (95% CI 75–80%), against 54% in 2016–19 (45–62%). The value-weighted median move climbed 45 places, against 9. Figure 2 shows the flow out of the Group of Five. Before 2020, about 1% of G5 stars who stayed in college moved to a Power program each year. The rate rose to 26% in 2023, 37% in 2024 and 59% in 2025 (95% CI 46–71%). Measured by value, the share of G5 proven talent exported to Power programs rose from 1.2% to 21.8%, and to 35.9% in 2025. FCS shows the same pattern from 2021, when its coverage becomes reliable: the share of FCS proven value moving up to FBS rose from 4.4% in 2021 to 28.6% in 2025.

The reverse flow is small. Proven value moving from Power to G5 programs averaged 65 units a season in 2022–25, against 590 moving up, a ratio of about nine to one (fifteen to one in 2025). Figure 3 maps the full flow between bands of the AVR board. Of all value that moved in 2022–25, 61% crossed into a higher band, 25% stayed within its band and 14% moved to a lower one. Before 2020 the entire flow was 12.6 SE a season; after 2021 it was 237.5 SE.

Table 2 summarizes the flows by band. In the pre-period, every band's net flow was within half a starter-equivalent unit per program per season. In the post-period, programs in the top 50 gained about 10–13 units (roughly one good starter) a season on net, programs ranked 51–100 gained about 4.5, and programs ranked 101–150 and below lost value. The top ten lost the smallest share of their returning value (10%), while programs ranked 51–150 lost more than a quarter.

The change is visible in who plays for Power programs. Players who had been at a G5 program the previous season took 0.3% of Power-program snaps in 2016 and 10.1% in 2025; former FCS players took 3.5%. Including transfers among Power programs, about 28% of Power snaps in 2025 went to players who had been elsewhere the year before.

6.3 Mobility growth across career stages

Figure 4 plots mobility rates for three groups of FBS regulars. Graduates moved at 3–6% a year before 2019 and at 26% on average in 2022–25. Undergraduates after their first or second season moved at under 0.5% before 2019 and at 18% after. Table 3 reports Eq. (2). Among regulars, graduate mobility rose 6.1-fold and undergraduate mobility 44-fold. The graduate increase matches 48% of the undergraduate increase on the log scale (κ = 0.48, 95% CI 0.33–0.62). Among all players, including backups, the rise was more uniform: 4.8-fold for graduates and 6.8-fold for undergraduates, so κ = 0.81 (0.68–0.97).

Under the proportional-growth assumption, the graduate increase corresponds to 48% of the undergraduate increase on a log scale among regulars and 81% among all players. These are exploratory decompositions. The shared component may include NIL, additional eligibility and changing transfer norms; it does not identify their individual effects.

The timing evidence points the same way. Almost all 2021 portal entries (98.5%) predate NIL's July 1 start, but entry dates do not establish when a move was completed or which incentives shaped it. In that year the share of G5 proven value moving to Power programs was 4.4%, against 2.0% in 2019. The large increases came afterwards, in step with the growth of NIL collectives: 8.6% in 2022, 18.6% in 2023, 24.0% in 2024 and 35.9% in 2025. A placebo split of the pre-period (2016–17 against 2018–19) moves the same measure by 1.2 points, against 20.6 points between periods. This timing is descriptive and does not isolate the causes of the increase.

6.4 Has talent concentrated at the top?

Recruiting data suggest dispersion. The share of blue-chip recruits on the rosters of the 25 programs with the most blue-chips fell from 75.7% to 61.2% (p = 0.030), consistent with Li and Derdenger (2025). At signing the change was smaller (81.1% to 77.2%), and the share who later transferred rose from 9% of the 2015 class to 48–62% of the 2019–2023 classes. Much of the roster dispersion therefore comes from blue-chips leaving after signing, not from where they signed.

Dispersion of recruits does not imply dispersion of ability. Figure 5 measures concentration on the full talent stock, including unproven players valued by their recruiting rating (§4.3). The top 10, 25 and 50 programs on the previous season's board held 9.5%, 22.5% and 42.9% of FBS talent in 2016–19 and 9.4%, 22.5% and 43.1% in 2022–25. The shares are similar. Crediting each transfer’s measured value to his previous program, the top 50 would have held 41.4% in 2022–25 and 39.4% in 2025. The accounting difference is 1.7 points to the top-50 share on average (95% CI 0.5–2.8) and 3.3 points in 2025. Elite programs lose players to the draft, to playing-time transfers and to graduation at a higher rate than before, and they use the transfer market to replace them.

The two pictures are reconciled by what moves. The blue-chips who leave elite programs are mostly backups who have not played; their expected contribution is small (an unproven four-star player is worth about 0.06 SE in a season, against roughly 0.5 SE for a proven starter in the top decile). The players who arrive at elite programs are proven. Unproven players made up 33% of transferred talent before 2020 and 22% after.

6.5 Team outcomes and competitive balance

Net transfer value is associated with changes in team strength. In 2022–25, the correlation between a program's net transfer value and its change in AVR rating, after removing regression to the mean, was 0.28; in 2016–19 it was 0.01 (Figure 6). The slope in Eq. (3) is 1.73 AVR points per SE in the post-period (95% CI 1.44–2.05). The pre-period slope is 0.21 but very imprecise (−2.25 to 2.78), and the difference between periods is not statistically distinguishable (one-sided bootstrap p = 0.13). The main change is not in what a unit of talent is worth but in how much of it moves: the standard deviation of net transfer value across programs rose from 0.18 SE to 1.38 SE.

Despite this, the overall balance of the sport has barely moved. On SP+, the standard deviation of team ratings (13.2 vs 13.0, p = 0.60), the gap between Power and G5 programs (15.4 vs 15.3 points, p = 0.83) and the correlation between talent and performance (0.65 vs 0.66) show no statistically distinguishable change in this sample. Team strength is somewhat less persistent from year to year (SP+ correlation 0.82 vs 0.76, p = 0.06), and swings of ten points or more became more common (19% vs 27% of programs, p = 0.08), but neither change is conventionally significant.

Table 4 looks at the top of the AVR board. Programs in the previous season's top ten were more likely to remain there (42.5% vs 60.0%, p = 0.08), while the number of new top-ten programs from outside the previous top 25 did not change (2.75 vs 2.5 a season). The rating gap between first and tenth narrowed from 9.8 to 6.7 points (p = 0.17). Nine programs finished in the top four in each period. These differences rest on four seasons a side and none reaches p < 0.05. Read together, they suggest a top group that is somewhat more stable in membership and somewhat less separated in strength.

6.6 Player outcomes

Performance after moving. Players who move up a level grade lower than matched stayers (Figure 8). G5 regulars who moved to a Power program graded 1.7 points below expectation in 2022–25 (95% CI −2.1 to −1.3; n = 425); FCS players moving to Power programs graded 1.9 points below (−2.5 to −1.3). Players moving down graded higher (Power to G5: +1.1, 0.0 to 2.0), and lateral moves show no difference. The same pattern, with similar magnitudes, appears in the small pre-period samples (G5 to Power −2.2; Power to G5 +2.3). We read this as a stable property of the grades, which are not adjusted for the strength of opposition, rather than as a decline in the players. About 58% of G5 regulars who moved up remained regulars, against 78% of those who stayed, and 19% of G5 stars who moved up graded as stars again, against 49% of those who stayed.

The draft. G5 regulars who moved to Power programs in 2022–24 were drafted at 14.4%, against 7.9% for matched G5 stayers (difference 6.4 points, 95% CI 2.9–9.9). FCS regulars who moved to Power programs were drafted at 9.4% against 2.3% (7.1 points, 3.9–10.8). Moves within or down from the Power category show the opposite: Power regulars who moved to G5 programs were drafted 20 points less often than matched Power stayers, consistent with these being moves for playing time by players a program chose not to keep. The share of drafted players who had transferred during college rose from 2–3% in 2016–17 to 34% in 2025 and 47% in 2026; the share of Power-program draftees who had played in the G5 or FCS rose from 3% to 21% (Figure 9).

6.7 What kind of transfer class works

Figure 10 decomposes incoming value in Eq. (3) for the post-period (1,043 program-seasons). Value earned at a program ranked the same or higher is associated with 4.5 AVR points per SE (95% CI 3.9–5.7); value earned at a lower-ranked program with 0.8 (0.3–1.2), about one-sixth as much. Each SE lost is associated with −1.7 points. Separating stars from other incoming players yields similar coefficients per unit (2.0 vs 1.5, overlapping intervals), so the composition of a class matters less than its total and its source. Holding value fixed, each additional arriving player is associated with +0.23 points (0.13–0.34), and we find no penalty for roster turnover as such. That last estimate may reflect depth, or programs under new coaches that rebuild and improve at the same time.

6.8 Has the flow levelled off?

The flow from the G5 to Power programs appears to have levelled off in 2026: 55.7% of G5 stars moved up (59.0% in 2025), the value share was 37.8% (35.9%), and portal entries were 4,474 (4,499). The FCS flow is still rising (31.8% of FCS proven value against 28.6%). Because 2026 counts include only players who have appeared in a game so far, the counts are lower bounds; the rates are less affected. Two seasons under direct revenue sharing are too few to judge its effect.

7. Robustness

Definition of talent. Table 5 re-estimates the main quantities under five alternatives to Eq. (1): grade thresholds of 55 and 65, counts of stars, counts of regulars, and snaps alone. Every definition gives the same picture. The share of talent that moves rises from about 2% to 20–29%; 70–82% of moved talent goes to a higher-ranked program; the top-50 share would be 1.3–3.3 points lower with transfers credited to origin programs; and the correlation between net transfer talent and rating change is 0.22–0.28.

Independent rating systems. The direction and concentration results use AVR ranks; the competitive-balance results use SP+. The team-outcome association is similar on both (Section 6.5 and an SP+ version with correlation 0.32 after removing regression to the mean).

Placebo. Splitting the pre-period in two produces changes an order of magnitude smaller than the change between periods for every flow measure we examined (for example, 1.2 points against 20.6 points for the G5 value share).

Count-based measures. Transfer counts alone show almost no association with team improvement (correlation 0.05), which is why value-weighting matters. This agrees with Pifer's (2026) finding that the picture reverses when transfers are weighted by talent.

Inference. Program-clustered and player-clustered bootstrap intervals are reported throughout. Season-level comparisons rest on four seasons a side, so we treat permutation p-values between 0.05 and 0.15 as suggestive.

8. Discussion

The findings are consistent with a secondary market for developed talent: substantially more proven value changes programs, while the top-50 share of measured talent remains similar. This pattern is compatible with Rottenberg’s (1956) invariance principle, but neither program valuations nor player payments are observed, so it is not a direct test of that mechanism.

Origin-credit accounting illustrates how transfers contribute to the observed distribution. It holds measured player value fixed and reassigns transfers to their previous programs; it does not model how recruiting, development or playing time would respond if transfers were unavailable. Likewise, similar competitive-balance estimates do not establish equivalence between periods.

Source context matters for the team-outcome association. The estimated coefficient per unit of incoming value from a lower-ranked program is about one-sixth that from a similar or higher-ranked program. This is a program-level association, not an individual-player conversion rule. Higher draft rates among upward movers may partly reflect selection into those moves.

9. Limitations

Identification. NIL, the transfer rule, the extra eligibility year, conference realignment, the expanded playoff and revenue sharing overlap in time. The graduate comparison separates the rule from the remaining factors only under a parallel-growth assumption, and the remaining factors cannot be separated from one another with these data. We have no data on NIL payments.

Measurement. PFF grades are not adjusted for opponent strength, and the value index in Eq. (1) is one of many possible summaries. Recruit-implied values for unproven players are averages by star rating and career stage. The roster link misses 9–22% of players a season, and its errors are likely concentrated among players with common names or mid-season roster changes.

Coverage. FCS grading before 2019 is sparse, so pre-period FCS flows are understated and excluded from cross-period comparisons. The 2019 roster has too few fifth-year players to estimate graduate mobility. Players who leave college football entirely are not observed after leaving, so all transfer shares are among players who remain. CFBD's portal list covers a subset of entrants.

Inference. Four seasons a side limit the power of season-level comparisons. The analysis is exploratory and not pre-registered, and many measures were examined; borderline results should be confirmed with additional seasons.

Selection. Matched comparisons of movers and stayers condition only on observed grades, origin category and unit.

10. Conclusion

Player mobility rose sharply after 2021, and most transferred proven value moved toward higher-ranked programs. The top-50 share of measured talent remained similar, while origin-credit accounting shows how transfers contribute to that share. Transfer value is associated with team improvement and upward movers have higher draft rates than matched stayers. These findings are consistent with a secondary market for developed talent, but do not establish the separate causal effects of NIL, eligibility reform or moving programs.

Data and code availability

PFF grades are licensed and cannot be redistributed. Roster, recruiting, portal and draft data are available from the CollegeFootballData.com API. SP+ ratings are published by their author. AVR ratings are published on this site. Analysis scripts and the derived tables behind every figure are available from the author on request.

Conflict of interest

The author maintains the Athletic Vector Rankings, which supply the Division I ranks used in Sections 6.2–6.7. The competitive-balance results in Section 6.5 use SP+, an independent rating, as a check.

References

1. Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much should we trust differences-in-differences estimates? Quarterly Journal of Economics, 119(1), 249–275.
2. Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2008). Bootstrap-based improvements for inference with clustered errors. Review of Economics and Statistics, 90(3), 414–427.
3. Dohrn, S. L., & Lopez, L. Y. P. (2022). NCAA transfer portal: Examining quarterback transfer outcomes in college football. Journal of Issues in Intercollegiate Athletics, 15(1), Article 22. scholarcommons.sc.edu/jiia/vol15/iss1/22
4. Fort, R., & Quirk, J. (1995). Cross-subsidization, incentives, and outcomes in professional team sports leagues. Journal of Economic Literature, 33(3), 1265–1299.
5. Gelman, A., & Loken, E. (2014). The statistical crisis in science. American Scientist, 102(6), 460–465.
6. Iacus, S. M., King, G., & Porro, G. (2012). Causal inference without balance checking: Coarsened exact matching. Political Analysis, 20(1), 1–24.
7. Li, I., & Derdenger, T. (2025). Does personalized pricing increase competition? Evidence from NIL in college football. Management Science. doi:10.1287/mnsc.2024.06423
8. NCAA. (2021, April 15). DI Council adopts new transfer legislation. ncaa.com
9. National Collegiate Athletic Association v. Alston, 594 U.S. 69 (2021).
10. Owens, Rennhoff, & Roach. (2025). The impact of name, image, and likeness contracts on student-athlete college choice. Applied Economics, 57(22), 2822–2838. doi:10.1080/00036846.2024.2331425
11. Pifer, N. D. (2026). Transfer networks and talent flow in the Football Bowl Subdivision transfer portal. Journal of Intercollegiate Sport, 19(2). doi:10.17161/jis.v19i2.24024
12. Rosen, S. (1981). The economics of superstars. American Economic Review, 71(5), 845–858.
13. Rottenberg, S. (1956). The baseball players' labor market. Journal of Political Economy, 64(3), 242–258.
14. Szymanski, S. (2003). The economic design of sporting contests. Journal of Economic Literature, 41(4), 1137–1187.

Appendix

A1. Coverage and linkage

A2. The 2019 graduate-rate artifact

In the roster data, the number of players in their fifth or later season since signing who appear on FBS rosters in consecutive seasons is 556 in 2018, 156 in 2019 and 332 in 2020. The 2019 shortfall is concentrated among non-movers, which inflates the measured 2019 graduate mobility rate to 35% (37% among regulars), against 3–7% in 2016–18 and 12–17% in 2020–21. Undergraduate rates show no break. We treat the 2019 graduate rate as a data artifact and use 2016–18 as the pre-period for Table 3. Including 2019 would raise the pre-period graduate rate and so increase θ, strengthening the role attributed to the rule.

A3. Transfer portal

A4. Definitions used in the full-talent measure

Expected current-season value for unproven players, in SE, by star rating and seasons since signing (0 to 4+): five-star 0.12–0.15; four-star 0.04–0.08; three-star 0.01–0.06; two-star or unrated 0.01–0.05. Proven players, by decile of prior value: 0.09 to 0.48 SE. The mapping is pooled across 2016–2025 so that a change in concentration cannot come from a change in the mapping.