How Daily Fantasy Sports Works
Daily fantasy sports contests let entrants build a lineup of real players under a salary cap and roster constraints, scoring points from those players' actual statistics in a set of games. Prizes depend on finishing position within the contest, so the best lineup depends on the contest format as well as on projected points.
The mechanics
Player pool and salaries. For a slate of games, the operator assigns every eligible player a salary intended to reflect expected production. Salaries are set before lineups lock and do not change with news, which is why late information matters so much.
Roster construction. Entrants select players to fill defined positions without exceeding the salary cap. Formats differ by sport and operator, including constraints such as a minimum number of teams represented or flexible positions.
Scoring. A fixed formula converts real statistics into fantasy points, for example points per yard, per reception, or per rebound, with bonuses in some formats. The formula is published and identical for all entrants.
Contest formats.
- Guaranteed prize pools, where a small share of entries win most of the prize money. Finishing near the top is what pays, so variance is valuable.
- Cash games, such as head-to-head or double-ups, where roughly half the field is paid. Consistency matters more than upside.
- Satellites and qualifiers, awarding entry to larger contests.
- Single-game contests, drawing from one match.
Locking and late swap. Lineups lock when games start, and some formats allow replacing players whose games have not yet begun. Availability news before lock is therefore decisive.
Why format changes the objective
In a contest paying half the field, maximising expected points is close to the right objective. In one paying the top one percent, a lineup with a high ceiling and a lower average can be better, because finishing in the middle pays the same as finishing last. The same projections lead to different optimal lineups under the two structures.
The data the problem needs
Projections. Expected fantasy points per player, built from usage, role, matchup, pace, and availability. This is a forecasting problem in its own right, and the scoring formula determines what matters: a format rewarding receptions values a different profile from one rewarding yardage.
Distributions, not just means. Because format determines the objective, a projection needs a variance estimate as well as a mean. Two players with identical projections and different ceilings are not interchangeable.
Salaries and constraints. The current slate's salaries, positions, and rules, which change per slate.
Availability and news. Injury designations, confirmed starters, expected minutes, and weather. Because salaries are fixed before news, late changes create the largest swings in value.
Correlation. Players in the same game and on the same team are correlated: a quarterback and his receivers rise together, and opposing defences fall together. Lineup construction is a correlated selection problem, which is why stacking and game selection are discussed so heavily.
Ownership projections. Estimates of what share of entries will select each player. In top-heavy formats, a strong player with very high ownership provides less differentiation than the raw projection suggests.
Optimisation. Selecting the best lineup under constraints is a classic integer programming problem, solvable with standard solvers, and the interesting part is the objective function rather than the solver.
A worked example of the format effect
Take two players at the same salary, one projected for 14 points with a narrow range and one projected for 13 with a much wider one. In a double-up paying half the field, the first is the better selection, because the target score is close to the field average and consistency clears it more often. In a large guaranteed prize pool, the second can be better, because clearing the field requires an unusual score and only the wider range reaches it. Neither projection is wrong. The contest structure decides which one is useful, and that is why a projection set without variance estimates is incomplete for anything but cash games.
Evaluating approaches honestly
Backtest against real slates. Historical salaries, player pools, and actual results, with projections built only from information available before lock.
Simulate the contest, not just the lineup. A lineup's value depends on the field. Simulating plausible opponent lineups from ownership estimates gives a far better estimate of finish distribution than scoring a lineup in isolation.
Measure the right outcome. For top-heavy formats, the distribution of finishes matters more than average points. A process with a higher average score can finish worse.
Account for rake and entry fees. Contest economics include the operator's cut, so beating the field by a small margin is not the same as profiting.
Watch sample size. Results in top-heavy formats are dominated by rare large finishes, so evaluation needs a very large number of contests to say anything.
Separate projection quality from construction. Evaluate projections against actual points independently of lineup performance, since a good projection model can be paired with poor construction and vice versa.
Legality and platform rules. Availability differs by jurisdiction and operator terms vary, including limits on entries per contest and on automated tools. Check both before building anything.
This page describes data and method and is not betting advice.
Frequently asked questions
- How does daily fantasy sports work?
- Entrants build a lineup of real players under a salary cap and positional constraints for a slate of games. A published formula converts those players' real statistics into fantasy points, and prizes depend on finishing position within the contest entered.
- What is the difference between cash games and guaranteed prize pools?
- Cash games pay roughly half the field, so consistent expected points matter most. Guaranteed prize pools concentrate prizes at the top, so lineups with higher ceilings and more variance can be preferable, since finishing mid-field pays the same as finishing last.
- Why does ownership matter in DFS?
- Because payouts depend on relative finish. In top-heavy contests, a player selected by a very large share of entries provides little differentiation even if the projection is strong, while a strong performance from a lightly selected player separates a lineup from the field.
- What data do you need for DFS modelling?
- Player projections with variance rather than just means, current salaries and roster constraints, availability and late news, correlation between players in the same game, and ownership projections. Contest simulation also needs plausible opponent lineups to estimate finish distributions. Salaries change every slate, so none of these inputs can be cached for long.