Possession Value Models: Football Analytics Guide for 2026

Picture Kevin De Bruyne picking up the ball just inside his own half. He scans, spots a run that hasn't started yet, and threads a weighted ball through three defenders. The finish is poor and the chance is gone. On the stats sheet, nothing happened: no assist, nothing to shout about. Yet that pass did exactly what a great pass should do, turning a harmless possession into a big chance. (It's a made-up scene, but you've watched a hundred like it.)

Possession value models, PVMs for short, exist to give that moment its credit. Instead of counting only goals and assists, they estimate how much every action changed a team's chances of scoring and conceding.

This guide walks through the main models (VAEP, EPV, OBV, g+ and xT), how clubs use them and, because this is a betting site, how much they're really worth to a punter. The short answer: a very useful lens, not a money machine.

Football has always loved a number. For decades, though, the numbers told only a sliver of the story. Goals, assists and clean sheets are the headline currency, yet they capture just the tip of the iceberg.

Take pass completion. A centre-back who completes 95% of his backward and sideways passes rates higher than a creative midfielder who keeps attempting the difficult ball between the lines. The percentage says nothing about how hard or how valuable each pass was.

Possession percentage has the same flaw. Dominating the ball in harmless areas means very little against a side that's happy to let you have it and hit you on the break.

And goals and assists reward only the final action in what is often a long, elaborate move. The forward who harries a defender into a mistake, the midfielder who recycles the ball under pressure, the defender whose positioning snuffs out a chance before it forms: all invisible on the traditional record.

Expected goals (xG) fixed part of the problem by measuring chance quality, but it still only looks at shots. So what about everything else that happens on a football pitch?

Rather than asking "Did this lead to a goal?", a PVM asks something more useful: "How did this action change the chance of scoring or conceding?"

Every action in a match, whether a pass, tackle, dribble, clearance or shot, changes the state of the game. A through ball into the box makes a goal more likely. A sloppy pass in midfield makes one against more likely. PVMs put a number on each of those shifts, then add them up for every player and team.

The basic arithmetic is a simple sum: the rise in your scoring chance, minus the rise in your conceding chance. Say a pass lifts your chance of scoring from 2% to 8% and cuts your chance of conceding from 1% to 0.5%. That's a gain of 0.06 goals plus a saving of 0.005, so the pass is worth about 0.065 goals. The numbers are invented for illustration, but the logic is how real models work.

Once you can do that for one pass, you can do it for everything. A last-ditch tackle that stops a certain goal is hugely valuable. A calm retention pass under pressure can be worth more than it looks. A speculative hoof that sails out for a throw-in carries negative value.

That is what makes these models so handy for the players the scoresheet forgets: holding midfielders, ball-playing centre-backs and pressing forwards, whose best work rarely shows up in the traditional record.

Several frameworks turn this idea into numbers, and each takes a different route. If you're serious about advanced football stats, these are the names to know.

VAEP (Valuing Actions by Estimating Probabilities)

Built by the DTAI research group at KU Leuven in Belgium, VAEP is one of the most influential academic models in football analytics. It uses machine learning to estimate how each action changes the probability of scoring and conceding over the actions that follow. Its strength is breadth: it can value anything from a simple pass to a dribble, and its open-source code has made it a foundation for plenty of later research. Individual action values are tiny, usually fractions of a goal.

EPV (Expected Possession Value)

Work led by Javier Fernández at FC Barcelona's analytics department produced EPV, which runs on tracking data showing where all 22 players are. That allows frame-by-frame analysis, including actions that never touch the ball: the dummy run that drags a defender away, or the movement that opens space for a team-mate. It is among the most detailed approaches, but the tracking infrastructure it needs puts it beyond most clubs.

OBV (On-Ball Value)

StatsBomb, now Hudl StatsBomb, introduced OBV in 2021. It is trained on expected goals rather than actual goals, which smooths out the randomness of finishing, and it splits the value between passer and receiver so nobody is counted twice.

g+ (Goals Added)

American Soccer Analysis created g+ for MLS and NWSL coverage. It values every touch and shares the credit for a pass between passer and receiver, because a good pass also needs a good first touch to come off.

xT (Expected Threat)

Karun Singh's xT is the easiest to picture. It carves the pitch into a grid, gives each zone a value based on how likely a goal is from there, and rewards any pass or carry that moves the ball into a more dangerous zone. Simple to explain and simple to build, it is a great place to start.

The academic roots might suggest PVMs live in research papers. They don't. Elite clubs have been folding these ideas into their work for years.

Recruitment

Liverpool under sporting director Michael Edwards, who held the role until 2022, are often held up as the poster child for analytics-led scouting. The aim was to find players whose contribution outstripped their goal and assist tallies: midfielders who moved the ball forward without the glamour stats, and defenders who prevented chances before they developed. Clubs rarely say how much rests on any single model, so take individual stories as informed reporting rather than a published method.

Match preparation

An opposition analyst might find that a rival generates most of its possession value from its right-back carrying the ball into midfield. The coaching staff can then build pressing triggers to shove play away from that zone. The same data shows which flank opponents favour, where they are vulnerable, and how often their set-piece routines create high-value positions.

Player evaluation

For scouts, PVMs give a fuller picture. A striker who isn't scoring but keeps creating high-value positions may be due a run. A defender whose traditional numbers look ordinary may be preventing plenty through positioning and anticipation. Full-backs who progress the ball from deep, Trent Alexander-Arnold during his Liverpool years being the obvious example, tend to shine on progressive-passing measures despite modest goal tallies.

Expected goals changed football analysis by measuring chance quality, but xG only captures the moments when shots happen. PVMs stretch the same thinking across the whole pitch.

Does wider mean better at predicting results? The evidence is mixed. A 2025 study in Frontiers in Sports and Active Living by Forcher and colleagues used three Bundesliga seasons (2022/23 to 2024/25). For pre-match prediction, EPV reached 58.3% accuracy against 55.6% for xG. After the match, though, xG was the stronger predictor (65.6% against 59.6%). The authors' reading: xG tells you more about how a team attacked once the game has been played, while pre-match EPV shows real promise for predicting what comes next.

For bettors, the practical upshot is that PVMs can flag teams creating high-quality possessions without converting them, who are candidates for positive regression, and teams running ahead of their underlying numbers. They can also expose tactical mismatches before the results do.

The betting market has been slower than elite clubs to embrace possession value, and the evidence that PVM data beats closing prices is thin. It is still a smart lens, and learning to apply football analytics to betting is a sensible way to start.

Spotting teams the results haven't caught up with

Teams that generate plenty of possession value but under-convert may be due better results, even if recent scorelines don't show it. The flip side: sides running ahead of their underlying numbers could be heading for a dip.

Player markets work the same way. A midfielder who keeps creating high-value positions without registering assists might be about to start, and a forward with strong possession value but poor finishing could be underpriced in scoring markets. Our guide to player shot markets shows how to put that thinking to work.

In-play

Real-time PVM tracking could reveal momentum shifts before they reach the scoreline, and might even help anticipate substitutions. That is one to test before you rely on it.

Pairing it with xG

The most effective analysis combines metrics. Run xG alongside possession value and the picture sharpens: strong xG with weak possession value suggests a team creating chances without really controlling games, while dominant possession value with poor xG conversion hints at quality that should come through. Above all, PVMs capture process, not just outcomes, and in a sport where variance bites hard, process is where the signal hides.

The next generation of PVMs pairs event data with player tracking, capturing all 22 players throughout the match. That means better value estimates and, crucially, credit for off-ball work.

Tracking data

The dummy run that drags a defender out of the danger zone can finally be quantified. Pressing that forces errors is measurable. Positioning that opens space for a team-mate shows up in the data. Javier Fernández's EPV work shows what is possible, breaking possession value into component models that value actions that never touch the ball.

AI and live analysis

Some clubs and data providers are experimenting with AI tools for live tactical and fatigue monitoring. The evidence on accuracy is still thin, so treat big claims about automated substitution advice or injury prediction as promising rather than proven.

Meanwhile, data is getting cheaper. Smaller clubs are gaining access to tracking technology, and sophisticated metrics are spreading well beyond the elite clubs that pioneered them.

For all their power, PVMs have real blind spots, and anyone leaning on them should know where they are.

What counts as a possession?

Nobody agrees. Should a cleared cross that comes back to the attacking team count as one possession or two? How many touches establish control? Should set pieces be treated separately? The choices matter: action-based windows tend to reward ball retention, while possession-based windows give more weight to forwards and shooting actions.

Off-ball work without tracking

Event-data models can't value what isn't in the event stream: off-ball movement, pressing intensity, positioning that creates space. Proxies introduce some bias, but they still beat having no context at all.

The black box

The most accurate models, tree-based ones like VAEP in particular, are hard to interpret. Why did the model give that action that value? Coaches won't trust numbers they can't argue with.

Who owns the data?

As tracking gets more detailed, so do the questions about who owns player movement and biometric data, and how sensitive fitness information is protected. The answers will shape how the field develops.

Possession value models are the biggest step forward in football analytics since expected goals. They credit the countless actions the scoresheet ignores, which explains player and team performance in ways goals and assists never could. For clubs they are one tool among several. For fans they answer why certain players and teams really succeed. For bettors they are a small, patient edge in a market still leaning on traditional numbers: real, but easily eaten by bookmaker margins and variance.

• Use PVMs alongside xG, not instead of it. The research suggests each wins at a different job.
• Know your source. Check which model and data provider you're looking at before comparing numbers; a VAEP figure and an OBV figure are not the same currency.
• Think seasons, not single matches. Sustained patterns mean something; one-game swings are mostly noise.
• Hunt for the mismatch. The best flags are teams creating plenty of value without the results to show for it, and the reverse.
• Keep the edge in perspective. Any advantage is small and has to beat the bookmaker's margin, so stake accordingly.

A pass that never becomes an assist isn't worthless. Possession value is how you prove it.

• Expected Threat (xT) in Football: Complete Analytics Guide
• VAEP Football Analytics: The Complete Guide to Valuing Every Action
• How to Build a Sports Betting Model: Complete Guide
• Home Advantage vs Team Quality: Advanced Football Betting Guide
• Over/Under 1.5 & 3.5 Goals: Advanced Goal Line Betting Strategies

18+. Please gamble responsibly. For free, confidential support visit BeGambleAware.org.