Fantasy basketball value stability and bust rates

Up: 2026-27 Fantasy Basketball Encyclopedia

Related: 2026-27 fantasy basketball rankings · 2026-27 fantasy basketball tiers · 2026-27 category rankings · 2026-27 points-league rankings · 2026-27 fantasy basketball player projections · 2026-27 NBA minutes projections · 2026-27 fantasy basketball projection methodology · 2026-27 fantasy basketball injury and availability ledger · 2026-27 sleepers, breakouts, and fades · 2026-27 fantasy basketball evidence and confidence policy · 2026-27 projection uncertainty ledger · How NBA injury reports should change fantasy decisions

The identity that governs everything

Season value decomposes exactly:

\[V_{\text{total}} = v_{\text{pg}} \times G\]

where \(v_{\text{pg}}\) is per-game fantasy value and \(G\) is games played. Realized value relative to expectation is therefore a product of two independent-ish ratios:

\[\frac{V}{\bar V} = \underbrace{\frac{v_{\text{pg}}}{\bar v_{\text{pg}}}}_{\text{skill/role term}} \times \underbrace{\frac{G}{\bar G}}_{\text{availability term}}\]

This is not a modelling choice. It is arithmetic, and it explains the asymmetry that follows. Rate declines are bounded and gradual; availability collapses are unbounded and instantaneous.

Worked comparison. Take a player you drafted expecting 62 games at a top-15 rate.

A 20% rate collapse — which in practice is a bad, narrative-generating season — costs you 10% of expected value. A 42-game absence at unchanged rate costs 68%. The availability failure is roughly seven times larger. This is the single most important number in the note.

A hand-audited season: RotoWire's preseason 2025-26 top 30

The cleanest verifiable exercise available at this cutoff is to take a published, timestamped preseason board and check outcomes one by one. RotoWire's 2025-26 preseason Top 200 is such a board. Its top 30 opened Jokić, Wembanyama, Gilgeous-Alexander, Dončić, Cunningham, Edwards, Antetokounmpo, Davis, Young, Towns.

Verified outcomes for the names that failed by a wide margin:

That is 4 catastrophic outcomes in a published top 30 — 13% — and every one of the four is an availability event. Not one is a case of a player whose per-game production simply stopped being good.

The near-misses reinforce the point rather than complicating it. Cade Cunningham was RotoWire's No. 5 and played 60 games before a collapsed lung (Yahoo) — a GP ratio near 0.91, an annoyance, not a bust. Jokić missed 17 games and still finished as the top per-game nine-category player (NBA.com final top 150). The rate held in both cases.

Honesty about scope. This is one season, one board, and I verified outcomes for the four large failures rather than computing a full 30-player rank delta. I could not locate a published multi-season fantasy-basketball hit-rate study of the kind that exists in fantasy football; the widely-circulated "bust rate" tables such as RotoWire's pick-slot analysis measure NBA draft slots against career VORP, which is a different question entirely. Treat 13% as one honest data point, not an established base rate.

The league-level availability regime is itself volatile

Individual risk sits on top of a league-wide condition that swings hard year to year, and this is verifiable:

A star-tier player's expected games swings from about \(0.795 \times 82 \approx 65\) to \(0.675 \times 82 \approx 55\) purely from league conditions — a 15% haircut on every elite asset simultaneously, before any individual medical history. Do not extrapolate the 2023-24 improvement; 2025-26 erased it.

Why per-game value is far more stable than total value

Take logs of the identity: \(\log V = \log v + \log G\). If the two terms are approximately independent, year-over-year correlation of total value is a variance-weighted blend:

\[\rho_V \approx \frac{\sigma_v^2 \rho_v + \sigma_G^2 \rho_G}{\sigma_v^2 + \sigma_G^2}\]

These inputs are estimates, clearly labelled as such — I could not verify published values. For established rotation players, plausible magnitudes are \(\sigma_{\log v} \approx 0.25\) with \(\rho_v \approx 0.80\) (roles and rates persist), and \(\sigma_{\log G} \approx 0.22\) with \(\rho_G \approx 0.35\) (games played is weakly autocorrelated — injury history predicts, but far less than people assume). Then:

\[\rho_V \approx \frac{(0.0625)(0.80) + (0.0484)(0.35)}{0.0625 + 0.0484} = \frac{0.0500 + 0.0169}{0.1109} \approx 0.60\]

Per-game rank carries forward at roughly 0.80; total-value rank at roughly 0.60. The availability term supplies ~44% of the variance but transmits only ~35% of it. The gap is irreducible. No better projection system closes it, because the missing information does not exist in August.

This is the mathematical justification for the encyclopedia's board contract in 2026-27 fantasy basketball rankings: rank per active game, keep games played as a separate visible column, and refuse to silently multiply the two.

Stability by player type

The critical distinction inside "low stability" is when the uncertainty resolves. A minutes question resolves in preseason, before your draft or immediately after it, and you can act. A hamstring question resolves in February, when you can do nothing. Ambiguous-role risk is therefore priceable; recurrence risk is not.

Late-round picks: you cannot win the league in Round 1, but you can lose it

2025-26 supplies concrete, verified cases of end-of-season starters acquired at or beyond pick 100:

That is at least three top-60 finishers from ADP >100 or undrafted, from a single non-exhaustive article. The true count of end-of-season top-50 players drafted outside the top 100 is plausibly 8–12 in a typical season — an estimate, not a verified figure — driven mechanically by 30 teams' worth of injury-created minutes.

But here is the asymmetry that matters. Assume each late slot has roughly a 5% chance of returning top-50 value. With four late picks plus ~20 in-season adds, you get ~24 draws:

\[P(\text{at least one top-50 hit}) = 1 - 0.95^{24} \approx 1 - 0.29 = 0.71\]

Roughly 70% of managers in your league will find at least one. Late-round hitting is table stakes, not an edge. Meanwhile your Round 1 pick busts at something like 13% and there is no second draw. You cannot differentiate yourself by winning the late rounds, because most of the room wins them too. You absolutely can eliminate yourself by losing Round 1.

The corollary: pay for availability at the top of the draft and take variance at the bottom, never the reverse.

Format note. In nine-cat, the late-round hit profile is a scarce-category specialist — Wallace's steals, Knueppel's threes — because one roster slot can win a category outright. In points leagues, value is nearly proportional to minutes × usage, category scarcity is worth zero, and late hits are simply minutes-volume players. See 2026-27 category rankings and 2026-27 points-league rankings. Points leagues make the GP term dominate even harder, because there is no way to win a week with a specialist.

What this implies for 2026-27

Reading the 2026-27 fantasy basketball rankings board against these base rates, the top-50 names carrying the most concentrated bust risk are the ones where two independent failure modes multiply:

The arithmetic of concentration. If the 13% top-30 rate holds, expect ~4 catastrophic outcomes in the top 30. If even three of those four come from the seven explicitly flagged names above, the flagged group busts near 43% and the unflagged group near 4% — an order-of-magnitude difference. That, not any individual opinion about a knee, is why the flags belong on the board rather than baked silently into a rank.

On the other side, the mid-round names in 2026-27 sleepers, breakouts, and fades with the widest upside share one property: their thesis is a minutes thesis, not a rate thesis. Payton Pritchard (Rounds 5-7, post-Jaylen Brown creation), Reed Sheppard (60-90, VanVleet rehab), Ausar Thompson (Rounds 5-7, projected 2.0 STL / 1.0 BLK), Cedric Coward (80-120), Ryan Rollins (70-105, already proven at top-60), Day'Ron Sharpe (85-125), and Collin Murray-Boyles (90-130) all resolve on an observable, October-dated signal: preseason minutes and closing lineups.

That is the practical conclusion. Buy resolvable uncertainty cheaply in the middle rounds; refuse to buy unresolvable uncertainty expensively in the first two. Minutes questions get answered before your season starts. Medical questions get answered in the standings.

Open questions

• No multi-season, multi-format fantasy-basketball hit-rate study appears to exist publicly. The 13% figure here is one hand-audited season against one published board. Building a real three-to-five-season ADP-vs-finish dataset would be the highest-value quantitative project in this encyclopedia, and it is tractable with historical ADP plus end-of-season nine-cat ranks.
• The \(\rho_v \approx 0.80\) / \(\rho_G \approx 0.35\) inputs to the correlation decomposition are reasoned estimates, not measured values. They should be computed directly from historical per-game z-scores and games played before anyone treats the 0.80-vs-0.60 spread as precise.
• Whether the 2025-26 availability collapse (67.5% star participation) is a regime change or a single bad draw is unresolved. It matters enormously for how much to discount every elite asset in 2026-27, and one more season of data will not settle it.
• The 5%-per-late-slot hit rate used in the \(1 - 0.95^{24}\) calculation is an assumption. It almost certainly varies by format, roster size, and how aggressively a manager churns the waiver wire — and the churn rate is probably the larger lever.