Possession Value Models Explained
Possession value models are one of the most important advances in modern football analytics because they measure what happens before the shot. Expected goals tells us the quality of a chance once a shot has been taken. Possession value asks a more difficult question: how much did each action increase or decrease the team’s likelihood of eventually creating a valuable attacking outcome?
That matters because football is not a sport of isolated shots. A high-quality chance may be created by a disguised pass through midfield, a centre-back breaking the first press, a winger receiving on the blind side, or a patient recycling action that moves the defence out of shape. Traditional metrics often reward the final action and ignore the quieter contribution that made the chance possible.
Possession value models try to close that gap. They assign value to passes, carries, receptions, dribbles, recoveries and other possession actions by estimating how the state of the attack changes. The core idea is simple: if an action moves the ball from a low-threat situation to a higher-threat situation, it has positive possession value. If it moves the ball into pressure, loses territory, or increases turnover risk, it may have negative value.
The details, however, are where the model becomes powerful.
What Is a Possession Value Model?
A possession value model estimates the expected attacking value of a team’s possession at different moments. Instead of waiting for a shot to occur, it evaluates the possession while it is still developing.
Imagine a midfielder receives the ball just inside his own half. At that moment, the possession has some probability of leading to a goal, some probability of ending in a low-value shot, and some probability of breaking down. If he plays a vertical pass into the half-space, the attacking team may now be in a more dangerous state. The model compares the value before the pass with the value after the pass.
The action value is usually expressed as:
the value of the possession after the action minus the value of the possession before the action.
If the value rises, the player has improved the possession. If it falls, the action has reduced the attacking threat.
This is why possession value models are sometimes described as “pre-shot” models. They are not replacing expected goals. They are extending the logic of expected goals backwards through the possession chain.
Why Expected Goals Alone Is Not Enough
Expected goals, or xG, measures the quality of shots based on features such as location, angle, body part and type of chance. It is excellent for evaluating finishing opportunities and team chance quality. But football contains many valuable actions that do not appear in an xG table.
Consider three players involved in a goal:
A centre-back breaks the press with a pass into midfield.
A midfielder turns and carries the ball through the centre.
A winger squares the ball for a striker to shoot from close range.
The striker receives the xG. The winger may receive an assist. But the centre-back and midfielder may have done the hardest tactical work: moving the ball through the most congested part of the pitch and turning a defensive possession into an attacking one.
Possession value models help distribute credit more fairly. They do not simply ask who shot or who assisted. They ask who increased the probability of a valuable attacking outcome.
This is especially useful for evaluating players whose contribution is structural rather than spectacular: deep midfielders, ball-playing defenders, inverted full-backs, press-resistant central midfielders and wide players who progress the ball before the final pass.
How Possession Value Is Calculated
Most possession value models begin with the same broad framework: define the state of the possession, estimate the value of that state, then measure how actions change that value.
1. Defining the Game State
The game state is a description of the situation before and after an action. A basic model may use only ball location. A more advanced model can include context such as:
match state
time remaining
type of action
body part
pressure
speed of attack
pass distance and angle
possession sequence information
defensive shape, if tracking data is available
player and team tendencies, where appropriate
A pass from the centre circle to the right wing is not always equally valuable. If the winger receives facing goal with space to attack, the value is different from receiving while trapped on the touchline under pressure. The richer the state description, the better the model can distinguish between superficially similar actions.
2. Estimating the Value of a State
Once the model understands the state, it estimates the expected value from that point. Different model families define this value differently.
Some models estimate the probability that the possession will lead to a goal. Others estimate the probability of creating a shot, or the expected goals value of the next shot. More complete approaches consider both attacking upside and defensive downside: if a risky central pass is intercepted, the opponent may counterattack from a dangerous area.
A simple example helps. Suppose a team has the ball near the halfway line. The possession may have a low expected value because it is far from goal. A successful line-breaking pass into Zone 14, just outside the penalty area, increases the probability of creating a shot. The value of the possession rises. The pass receives positive possession value.
If the same midfielder instead plays a loose square pass that leaves a teammate under pressure, the immediate ball location may not look terrible, but the risk of losing possession may rise. A strong possession value model should recognise that the action has not improved the attack in a meaningful way.
3. Assigning Value to Actions
The action value is the change in possession value caused by the action.
For a completed pass, this is often measured by comparing the estimated value at the pass origin with the estimated value at the reception location and context. For a carry, it compares the value at the start and end of the carry. For a failed action, the model accounts for the resulting turnover and the value handed to the opponent.
This is crucial. A player who constantly attempts ambitious forward passes may generate positive value when they work, but costly negative value when they fail. A possession value model should capture both sides of the trade-off.
Football decisions are not only about progression. They are about risk-adjusted progression.
Practical Examples of Possession Value in Football
Possession value becomes clearer when applied to familiar tactical situations.
The Safe Centre-Back Pass
A centre-back passes sideways to his defensive partner under no pressure. The pass is completed, but the ball remains in a similar area with a similar attacking outlook. The possession value barely changes.
This does not mean the pass is useless. It may be part of a larger structure, helping the team draw pressure or reset circulation. But as an individual action, its immediate possession value is likely to be small unless it meaningfully changes the opponent’s defensive shape.
The Line-Breaking Midfield Pass
A defensive midfielder receives from the back line and plays through pressure into an attacking midfielder between the lines. The ball moves from a relatively safe but low-threat area into a central pocket closer to goal.
This action usually increases possession value because it removes defenders from the play and gives the receiving player access to forward options. Even if no shot follows immediately, the action has improved the attacking position.
The Wide Carry That Creates Crossing Access
A winger receives near the halfway line and carries the ball 25 metres into the final third. The carry may not directly create a shot, but it changes the possession from a build-up state into an advanced wide attacking state.
Whether the action receives high value depends on the context. If the winger carries into space with teammates arriving in the box, the value rises. If the carry ends with the player isolated near the corner flag and no support, the model may assign only modest value.
The Risky Central Turnover
A midfielder attempts a difficult pass through the centre and loses the ball. The attacking value of his team’s possession drops to zero, but the opponent may also gain a dangerous transition opportunity.
This is where possession value models become more realistic than simple progression metrics. Moving the ball forward is not automatically good. Losing it in central areas can be more damaging than losing it near the opposition corner flag.
Possession Value Versus Progressive Passes
Progressive passes are useful, but they are not the same as possession value. A progressive pass usually measures whether the ball has moved significantly closer to goal. Possession value measures whether the action has improved the expected outcome of the possession.
A long diagonal from centre-back to winger may be progressive, but if the winger is immediately surrounded and forced backwards, the actual value may be limited. Conversely, a short pass into a central midfielder between two lines may travel only eight metres but dramatically increase attacking potential.
This distinction is essential for player evaluation. A model that only counts progression may overrate players who move the ball forward without considering reception quality, pressure, structure or next-action potential. Possession value is more sensitive to what the action actually does to the possession.
Different Types of Possession Value Models
There is no single universal possession value model. Most approaches fall into a few broad categories.
Zone-Based Models
The simplest models divide the pitch into zones and estimate the value of possessing the ball in each area. Moving the ball from a low-value zone to a high-value zone creates positive value.
These models are easy to understand and useful for education, but they can be blunt. They may treat all possessions in the same zone as similar, even though pressure, body orientation and defensive positioning can change the situation dramatically.
Action-Based Models
Action-based models evaluate specific events: passes, carries, dribbles, shots and defensive actions. They can account for success probability and the value of the resulting state.
These models are better suited to player analysis because they assign value directly to football actions rather than only to locations.
Possession-Outcome Models
Some models estimate the likelihood that a possession will lead to a shot, a goal, or a chance of a certain quality within a defined sequence. They connect every action to the eventual outcome of the possession.
The challenge is avoiding excessive credit for actions that were only loosely connected to the final shot. A sideways pass 15 actions before a goal should not receive the same treatment as a pass that breaks the defensive line. Good models therefore focus on state changes rather than simply distributing credit across all actions.
How Football Data Lab Applies Possession Value Models
In Football Data Lab’s modelling pipeline, possession value is treated as a bridge between event analysis and prediction modelling. The aim is not merely to label a pass as “good” or “bad”, but to estimate how each action changes the attacking and defensive expectation of the possession.
The process begins by converting event data into possession chains. Each chain is segmented into ordered actions: receptions, passes, carries, duels, shots, turnovers and set-piece events where relevant. Every action is stored with its pre-action and post-action state, including pitch location, action type, possession phase, direction of play, sequence length and contextual match information.
The model then estimates two connected values at each point in the chain.
The first is attacking possession value: the expected future shot value for the team in possession. This is anchored to expected goals at the terminal point. If a possession ends in a shot, the shot’s xG becomes the realised attacking endpoint. If it ends without a shot, the model learns from similar historical states to estimate the probability and likely quality of a future shot from that situation.
The second is concession risk: the value potentially handed to the opponent if the action fails or if the possession state becomes unstable. This matters because a pass into the centre of the pitch can carry both upside and downside. A model that only rewards successful progression will miss the cost of exposing the team to transition.
For each action, Football Data Lab calculates a simplified net possession value:
post-action attacking value minus pre-action attacking value, adjusted for turnover and opponent transition risk.
That adjustment is important. A full-back pass into midfield may look valuable because it moves the ball centrally, but if the receiving player is under heavy pressure and the next state has a high turnover profile, the net value is reduced. A backwards pass after drawing pressure may score better than expected if it preserves possession and creates a cleaner switch on the next action.
Football Data Lab also separates passer value from receiver context where the data allows. A perfectly weighted pass into a dangerous pocket and a poor reception in the same zone should not always be treated identically. The pipeline therefore evaluates the change in state created by the pass and, separately, the value of the receiver’s next action. This avoids over-crediting the final passer in moves where the receiver creates most of the advantage with the first touch or carry.
The possession value output is then aggregated at player, team and match level. For players, it helps identify repeatable contribution profiles: who adds value through central progression, who adds value through wide entries, who preserves value under pressure, and who creates high-upside possessions at the cost of volatility. For teams, it helps measure whether territorial control is actually becoming chance creation or merely sterile possession.
The final layer connects possession value to match prediction. Expected goals remains the main currency for estimating scoring strength because goals come from chances, not from possession value directly. However, possession value helps explain the process behind those xG numbers. A team consistently generating high pre-shot possession value may be building attacks that are likely to produce future xG, even if the most recent match did not contain many shots. In forecasting, xG feeds the goal expectation component of match predictions, while possession value informs the assessment of chance creation sustainability, build-up quality and tactical matchup effects.
A simplified example from the pipeline would treat two teams with similar recent xG differently if their possession value profiles diverge. Team A may rely on a small number of low-repeatability moments, such as long shots or broken-play chances. Team B may regularly move the ball into high-value central and wide zones before the shot occurs. Even if their recent xG totals are similar, the second profile may be considered more structurally reliable because the underlying possession process is healthier.
Why Possession Value Is Useful for Player Recruitment
Possession value models are particularly valuable in recruitment because they reveal skills that traditional data can miss.
A midfielder at a mid-table club may not record many assists because his team lacks elite finishers. But if he consistently increases possession value through line-breaking passes and carries, he may be creating the conditions for chances even when teammates fail to convert them into shots.
A full-back may have modest crossing numbers but excellent value through underlapping carries and passes into the box. A centre-back may appear conservative by raw pass distance but add value by finding midfielders in pressure-resistant positions.
Recruitment departments can use possession value profiles to compare players across teams with different styles. The key is not simply ranking players by total value. It is understanding how that value is produced.
Does the player progress centrally or wide?
Do they add value against settled blocks or mainly in transition?
Is their value volume-driven or efficiency-driven?
Do they take high-risk actions that fit the recruiting team’s structure?
Can their current role translate to a different tactical environment?
A possession value model does not answer all of these questions alone, but it gives analysts a sharper starting point.
Common Misinterpretations of Possession Value
Possession value is powerful, but it can be misread.
The first mistake is treating it as a perfect measure of player quality. It is not. It measures contribution through observed actions in a specific tactical context. A player’s role, team structure and available passing options all influence their output.
The second mistake is assuming high possession value always means attacking brilliance. Some players accumulate value because they receive the ball in favourable positions created by others. That is why role analysis and video review remain essential.
The third mistake is ignoring defensive trade-offs. A player who creates high attacking value but loses the ball in dangerous areas may be less useful than the headline number suggests. Net value matters more than attacking value alone.
The fourth mistake is comparing positions too directly. Centre-backs, holding midfielders, wingers and strikers interact with possession value in very different ways. The best comparisons are role-specific rather than universal.
The Limitations of Possession Value Models
No possession value model sees the whole game perfectly. Event data often lacks information about off-ball movement, defensive spacing and body orientation. Tracking data improves this, but even then the model must simplify a complex tactical environment.
There is also the issue of causality. If a pass into the box leads to a shot, the pass may receive value. But perhaps the real reason the chance existed was a decoy run that dragged a defender away. Possession value models can under-credit off-ball actions because they are harder to observe in standard event data.
Another limitation is tactical intent. A team protecting a lead may choose lower-risk possession. A model must be careful not to punish every backwards pass if the strategic objective is control rather than chance creation.
Finally, possession value can be noisy in small samples. A few successful high-value actions can distort early conclusions. The most reliable insights come from repeated patterns across matches, roles and tactical contexts.
FAQ: Possession Value Models
What does a possession value model measure?
A possession value model measures how much an action changes the expected value of a team’s possession. It estimates whether a pass, carry, dribble or other action improves the likelihood of creating a valuable attacking outcome.
How is possession value different from expected goals?
Expected goals evaluates shot quality. Possession value evaluates actions before the shot. xG tells us how good the chance was; possession value helps explain how the team moved the ball into a position to create that chance.
Can possession value evaluate defensive players?
Yes. Ball-playing centre-backs and full-backs often contribute by progressing possession from deeper areas. Possession value can identify defenders who break pressure, access midfield, or create stable attacking platforms.
Does a backwards pass always have negative possession value?
No. A backwards pass can be valuable if it helps escape pressure, preserve possession, or open a better attacking route. The model should evaluate the change in possession state, not simply the direction of the pass.
Are possession value models useful for match prediction?
They can be. Expected goals remains central for estimating scoring strength, but possession value helps assess whether a team’s chance creation process is sustainable. It provides context behind xG by measuring the quality of ball progression before shots occur.
Conclusion
Possession value models give football analysts a way to measure the hidden work before the shot. By estimating how each action changes the expected outcome of a possession, they reveal the value of progression, pressure resistance, circulation and risk management.
The best use of possession value is not as a single ranking metric, but as a diagnostic tool. It helps explain how teams create chances, how players contribute within roles, and whether attacking performance is built on repeatable possession quality or isolated moments. In modern football analytics, that makes possession value essential for understanding the game between the passes and the shot.
For probabilities on today's full fixture list, see our live match predictions.