Possession Value Models: Football Analytics Guide for 2026

Kevin De Bruyne receives the ball near the halfway line. He scans the field, spots a run that hasn't happened yet, and plays a perfectly weighted through ball that splits three defenders. The chance gets missed, but De Bruyne's contribution doesn't disappear from the record.

According to possession value models, that pass - the one that didn't produce a goal or an assist - increased Manchester City's probability of scoring by roughly 12%. In the language of modern football analytics, De Bruyne just added 0.12 expected goals through an action that traditional statistics would have ignored entirely.

This is what possession value models (PVMs) do: capture the 99% of football actions that don't end in shots but determine who wins matches. For clubs, analysts, and bettors who have embraced them, these models are revealing truths about the game that were previously invisible.

Football has always been a numbers game. But for decades, those numbers told only a fraction of the story. Goals, assists, clean sheets - these are the currency of football discourse, yet they capture just the tip of the iceberg.

Take passing percentage. It tells you nothing about the difficulty or value of those passes. A centre-back completing 95% of backward and sideways passes rates higher than a creative midfielder attempting difficult through balls.

Or possession percentage. It ignores what teams actually do with the ball. Dominating possession in harmless areas means little against a counter-attacking side content to let you have it.

Goals and assists? They reward only the final action in often elaborate sequences. The pressing forward who forces a mistake, the midfielder who recycles possession under pressure, the defender whose positioning prevents a chance before it develops - all invisible in the traditional record.

The analytics revolution that brought us Expected Goals (xG) solved part of this problem by measuring chance quality. But Expected Goals (xG) still focuses on shots. What about everything else that happens on a football pitch?

Possession value models represent a fundamental shift in how we think about football actions. Rather than asking "Did this lead to a goal?" they ask something more useful: "How did this action change the probability of scoring or conceding?"

The core concept is simple. Every action on a football pitch - every pass, tackle, dribble, clearance, or shot - alters the game state. A through ball into the box increases the chance of scoring. A misplaced pass in midfield increases the chance of conceding. PVMs assign numerical values to these changes.

If a midfielder's pass moves the ball from a position where scoring probability was 2% to one where it's 8%, while also reducing conceding probability from 1% to 0.5%, that pass is worth approximately 0.055 goals in value.

This framework lets analysts value everything. A last-ditch tackle that prevents a certain goal? Highly valuable. A simple retention pass under pressure? Surprisingly valuable. A speculative long ball that goes out for a throw-in? Negative value.

For the first time, we can quantify the contributions of defensive midfielders, ball-playing centre-backs, and pressing forwards whose work never appears on a traditional scoresheet.

Several frameworks have emerged to operationalize possession value, each with distinct approaches and philosophies. If you're serious about advanced football statistics, you'll want to understand VAEP, EPV, OBV, and xT.

VAEP (Valuing Actions by Estimating Probabilities)

Developed by the KU Leuven DTAI Lab in Belgium, VAEP is perhaps the most influential academic model in football analytics. It values over 36 different action types using machine learning, considering both the probability of scoring and conceding within the next 10 actions.

VAEP's strength is its comprehensiveness. It can value everything from a simple pass to a complex dribble, and its open-source nature has made it a foundation for further research. The model typically rates actions between -0.05 and +0.10 goals, with elite performers like Lionel Messi and Kevin De Bruyne consistently achieving +0.50 to +0.90 VAEP per 90 minutes.

EPV (Expected Possession Value)

FC Barcelona's analytics team, led by Javier Fernandez, developed EPV using tracking data that captures the position of all 22 players. This allows for frame-by-frame analysis of possession value, including actions that don't involve the ball.

EPV can value off-ball movements - the dummy run that drags a defender away, the positioning that creates space for a teammate. It's the gold standard for possession value analysis, but it requires expensive tracking infrastructure that puts it beyond the reach of most clubs.

OBV (On-Ball Value)

Hudl StatsBomb introduced OBV in 2021, trained on Expected Goals rather than actual goals. This smoothing approach reduces the volatility that comes from the randomness of finishing. OBV extends the possession window beyond immediate actions and splits value between passer and receiver to avoid double-counting.

g+ (Goals Added)

American Soccer Analysis created g+ for MLS and NWSL coverage, taking a possession-based approach that values all touches. It notably splits pass value between the passer and the receiver, recognizing that a brilliant pass requires a good first touch to complete.

xT (Expected Threat)

Karun Singh's xT model uses a grid-based approach to value ball progression. Each zone on the pitch has an inherent value based on the probability of scoring from there, and moving the ball between zones generates positive or negative value.

The academic roots of possession value models might suggest they're confined to research papers. Nothing could be further from the truth. Elite clubs have been quietly incorporating these frameworks into their operations for years.

The Liverpool Story

"Liverpool famously used possession value to help scout its squad that conquered Europe on a modest budget," reported FiveThirtyEight in their deep dive into the club's analytics revolution.

Under Sporting Director Michael Edwards, Liverpool built a recruitment strategy that identified players whose contributions exceeded their goal and assist tallies. They found midfielders who progressed the ball effectively without registering traditional statistics. They discovered defenders whose positioning prevented chances before they developed. The result? A Champions League-winning squad built by finding value where others weren't looking.

Tactical Analysis and Match Preparation

Coaches use PVM insights to understand opponents at a granular level. An opposition analyst might identify that an upcoming opponent generates most possession value through their right-back's carries into midfield. The coaching staff can then design pressing triggers that force the ball away from that zone.

The models reveal which flanks opponents favor for ball progression, where defensive vulnerabilities exist, and which players carry the highest risk/reward profiles. Set-piece routines can be analyzed for their effectiveness in creating high-value positions.

Player Evaluation Beyond Goals

For scouts and recruitment teams, PVMs offer a more complete picture of player contribution. A forward who isn't scoring but consistently creates high-value positions might be due for a goal-scoring run. A defender whose traditional statistics look ordinary might be preventing significant value through positioning and anticipation.

Full-backs like Trent Alexander-Arnold rank among the top contributors in possession value models despite rarely scoring. Their progressive passing from deep positions consistently creates dangerous situations. The numbers align with what attentive viewers see: Alexander-Arnold is among the most valuable players in world football, even if his goal contributions don't reflect it.

Expected Goals revolutionized football analysis by quantifying chance quality. But xG captures only the final third of the pitch - the moments when shots occur. PVMs extend this framework to every blade of grass.

| Aspect | Expected Goals (xG) | Possession Value Models |
|--------|---------------------|-------------------------||
| What it measures | Quality of scoring chances | Value of ALL actions on pitch |
| Actions valued | Shots only | Passes, dribbles, tackles, clearances |
| Defensive value | Limited (xG prevented) | Full defensive contribution |
| Build-up play | Not measured | Every action valued |
| Scope | Final third | Entire pitch |

The difference matters for evaluation and prediction. According to 2025 Bundesliga research published in Frontiers in Sports, EPV showed higher prediction performance (58.3% accuracy) compared to xG (55.6% accuracy) for pre-match scenarios. The additional context of how teams build play, not just how they finish chances, provides predictive power.

For bettors and analysts, this means PVMs can identify teams generating high-quality possessions without converting - prime candidates for positive regression. They can spot teams overperforming their underlying numbers and detect tactical mismatches before they manifest in results.

The betting market has been slower to adopt possession value models than elite clubs, but that's changing. Savvy bettors are finding edges in PVM data that traditional statistics miss. Learning how to apply football analytics to betting can provide a real competitive advantage.

Finding Value in Football Betting Markets

Teams that generate high possession value but underconvert present opportunities. Their underlying numbers suggest better results ahead, even if recent scoresheets don't reflect it. Conversely, teams overperforming their PVM metrics may be due for negative regression.

Player prop markets work similarly. A midfielder creating consistent high-value positions without registering assists is statistically likely to start producing. A forward with strong possession value but poor finishing might be undervalued in scoring markets.

Live Betting Applications with Real-Time Analytics

Real-time PVM tracking enables sophisticated in-play betting. Monitoring possession value during matches can reveal momentum shifts before they appear on the scoreline. Identifying tired players likely to be substituted provides edges in card and substitution markets.

The key insight here: PVMs capture process, not just outcomes. In a sport where variance significantly affects results, focusing on the process can separate signal from noise.

Combining xG and Possession Value for Better Predictions

The most effective betting analysis combines multiple metrics. Using Expected Goals (xG) alongside possession value models provides a more complete picture of team performance. A team with strong xG numbers but weak possession value might be creating chances without controlling matches. Dominant possession value with poor xG conversion suggests underlying quality that may improve.

Possession value models continue to evolve rapidly, driven by advances in artificial intelligence and the increasing availability of tracking data.

The Tracking Data Revolution

The next generation of PVMs combines event data with player tracking, capturing the position of all 22 players throughout the match. This enables more accurate value estimation and, crucially, allows analysts to value off-ball movements.

The dummy run that drags a defender away from the danger zone? Now quantifiable. The pressing intensity that forces errors? Measurable. The physical positioning that creates space for teammates? Visible in the data.

Javier Fernandez's EPV model demonstrates what's possible: breaking possession value into component models that assign value to actions that never touch the ball. The implications for tactical analysis are significant.

AI and Real-Time Analysis

By 2026 and beyond, expect artificial intelligence to provide real-time tactical recommendations during matches. Automated substitution suggestions based on fatigue and performance data. Predictive injury risk modeling using movement patterns.

The democratization of data is accelerating too. Smaller clubs are gaining access to tracking technology. Lower-tier leagues are implementing data infrastructure. Sophisticated metrics are spreading beyond the elite clubs that pioneered them.

For all their power, possession value models have significant limitations that analysts and bettors need to understand.

The Possession Definition Problem

There is no consensus on what constitutes a "possession." Should a cleared cross that returns to the attacking team count as one possession or two? How many touches establish control? Should set-pieces be treated separately?

These questions matter. Research from KU Leuven shows that action-based windows tend to rank defenders higher (rewarding ball retention), while possession-based windows give more weight to forwards and shooting actions.

Off-Ball Contributions Without Tracking

Current event-data models cannot value actions that don't appear in the event stream. Off-ball movement, pressing intensity, communication, leadership, physical positioning that creates space - all invisible without expensive tracking infrastructure.

As Jan Van Haaren, Club Brugge data scientist and VAEP co-developer, acknowledged: "There could be a slight bias when using proxies. But I still think it's better than using no context at all."

Model Transparency

The most accurate models - particularly tree-based approaches like VAEP - are difficult to interpret. Why did the model assign this specific value? How can coaches trust outputs they don't fully understand?

This "black box" problem becomes more acute as artificial intelligence drives model development. The balance between accuracy and interpretability remains an ongoing tension.

Data Ownership and Privacy

As tracking data becomes more sophisticated, questions of ownership and privacy intensify. Who owns player movement and biometric data? How is sensitive fitness data protected? These ethical considerations will shape how the field develops.

Possession value models represent the most significant advance in football analytics since Expected Goals. They capture the 99% of actions that traditional statistics ignore, revealing truths about player contribution and team performance that were previously invisible.

For elite clubs, PVMs have become essential tools for recruitment, tactical analysis, and match preparation. For bettors and analysts, they offer edges in markets still dominated by traditional metrics. For fans, they provide a richer understanding of why certain players and teams succeed.

The technology isn't perfect. The models disagree on fundamentals like possession definitions and shot valuation. They can't capture everything that happens on a pitch, and their complexity can make them difficult to trust.

But the direction is clear. As tracking data becomes ubiquitous and artificial intelligence advances, possession value models will only grow more sophisticated and more influential. The clubs, analysts, and bettors who master them now will have advantages that compound over time.

Kevin De Bruyne's through ball that didn't become an assist? It wasn't worthless. In the language of possession value, it was worth 0.12 goals. Understanding that distinction might just be the key to understanding modern football analytics.

• How to Use xG (Expected Goals) for Smarter Football Betting
• How to Build a Sports Betting Model: Complete Guide
• Player Shot Markets: How to Profit from Shots on Target
• Correct Score Betting Strategy: The Mathematical Approach
• European League Betting Guide: La Liga, Serie A & Bundesliga Value