Reading 248 V-League 2026-2026 Matches Through xG: Three Teams Living on Luck, One Team Being Misjudged

Hook — A Night at Thien Truong, a Shot Straight at the Goalkeeper

June 15, 2026, Thien Truong Stadium. Nam Dinh FC host Ha Noi FC. Minute 87, score 1-1. The visitors have taken 14 shots, xG at 2.41. The hosts have five shots, xG at just 0.78. The whole stand rises as a Ha Noi striker meets a long ball and goes one-on-one with goalkeeper Tran Nguyen Manh. The shot goes straight at the keeper. Three minutes later, Nam Dinh counter, the ball flies from the edge of the box into the far corner. 2-1.

I sit in row twelve of stand B, a battered notebook in hand. Eight years after the Hang Day xG shock, I still do not trust what my eyes see in the first 90 minutes. I trust what the model says after the match ends. That night at Thien Truong, the model said Ha Noi FC should have won by two. The table recorded something else.

That is why I reopened the entire 2026-2026 dataset. Not to find someone to blame. But to check where the gap between xG and goals sits, and who is paying for it.

Context — Where I Built This Model From

I started logging xG in Vietnam in 2026, after losing 180 million VND on a bet when Ha Noi FC drew Quang Nam FC 1-1 at Hang Day. In that match Ha Noi took 17 shots, xG 2.87, while the opponent had two shots with xG 0.94. Furious, I reviewed 112 V-League matches from round 1 to round 14, manually calculating xG for every shot. The result changed how I work: Ha Noi FC created plenty of chances but finished 23% below the league average. A month later, that data predicted their four-match losing streak.

In 2026-2026, I scaled up. I tracked 248 matches across the V-League and national cup system, logging every shot by coordinates, angle, type of preceding pass, number of defenders in the box, and match state at the moment of the shot. From that I computed xG per shot, aggregated into match xG, and compared it with actual goals. I also logged PPDA — passes allowed per defensive action — to measure pressing intensity, and distance covered to measure physical load.

After the COVID-19 experience, when football returned to empty stadiums, I abandoned the idea of "absolute data." I designed what I call a "context coefficient": adjusting xG for whether the ground has fans, the weather, the away team's travel distance, and fixture congestion within a month. Without a context coefficient, every table is a form of self-deception.

The 2026-2026 season brought something particular. The rise of modest-budget clubs into the top group, the slowing of traditional giants, and the arrival of several foreign coaches with data-driven philosophies. Emotionally, this was the season of national-team belief, of ASEAN Cup and qualifier nights. Tactically, it was the season in which the gap between xG and goals spoke loudest in my eight years of tracking.

Core — The Evidence Chain from 248 Matches

The first thing I found when comparing the points table with the xG table: three teams in the top six sit four to seven places higher than their xG ranking. This is a large gap by Asian standards. In the European leagues I cross-checked, the average gap between points position and xG position after 20 rounds is usually only two to three places. In the V-League this season, the average deviation reached 4.3 places. The cause is not my xG formula. It is what I call "accumulated random residue."

Let me start with a specific club. I will not name them yet, because I want you to read the numbers before you read the name.

Team A scored 34 goals after 20 rounds, but xG stood at just 26.4. A positive differential of 7.6 goals. Across the 248 matches I tracked, this is the third-highest positive differential in the league. Analyzing their shots in detail, I noted two things. First, 11 of the 34 goals came from outside the box, where average xG per shot is only 0.05. That is one goal per 20 long-range shots. Team A scored 11 goals from 71 shots outside the box — a 15.5% conversion rate, three times the expectation. Second, nine of the 34 goals came from fast counters after an opponent turnover, and six of those were shots from inside 12 meters in situations my model rated at only 0.14 to 0.19 xG.

There is nothing wrong with scoring. But when a team scores 29% above xG, one of two things is true: either they possess a truly elite finishing group, or they are riding a lucky streak that can reverse at any moment. Across 248 matches I found evidence for both, but the odds lean toward the second.

Historical cross-check: in the 2026 season, a club had a similar positive differential of 8.1 goals after 20 rounds, then dropped nine points in their final six matches and lost their continental spot. In 2026, another club carried a positive differential of 6.8 goals after 18 rounds, then finished the season at minus 1.4 after 26 rounds. Nothing guarantees Team A repeats this. But the probability of mean reversion for clubs with positive differentials above 25% over the last seven V-League seasons is 71%.

Team B is different. Team B scored 28 goals after 20 rounds, xG 31.2. A negative differential of 3.2 goals. On the surface, an "inefficient" side. But when I broke it down by shot type, the picture shifted. Team B scored 12 goals from open play inside the box with an average xG per shot of 0.28 or higher. That is a good number. Their problem lay elsewhere: nine of their 20 matches were played on pitches that failed standards due to heavy rain, and I had to apply a 0.91 adjustment to both teams' xG in those matches. After adjustment, Team B's true xG fell to 28.4, close to their 28 actual goals. This is the blind spot of every data model in Vietnam: infrastructure has never been factored into the equation, and it can distort the picture by up to 10%.

{{IMAGE:1}}

Team C is the team that made me write this piece. Team C scored 22 goals after 20 rounds, xG 30.8. A negative differential of 8.8 goals — the highest in the league. On the table they sit tenth. On the xG table they sit fifth. In four of their last five matches, Team C took more than 14 shots, averaged 1.95 xG per match, and scored one goal in total. I rewatched every shot of Team C across those four matches. Fifty-eight shots in total. Thirty-four hit the frame or were blocked inside the box. Six struck the post or crossbar. Eight were denied by the keeper from inside 10 meters. Of those 58 shots, only one was taken in a comfortable posture — the ball bounced into the correct stride, no defender marking. That shot went in.

That is data. No emotion in it. When a team takes 58 shots and only once finishes in the correct posture across four matches, the cause is not mentality. The cause is attacking structure: Team C creates chances, but those chances always arrive in a chaotic state, with no one standing in a clean finishing position.

I returned to the footage. In minute 63 against Team D, Team C countered four against three. The ball came in from the right. The first runner into the box stood between two defenders. The last runner drifted toward the near post. No one occupied the central position. The finisher had to receive with his back to goal. He took 1.2 seconds to turn. In those 1.2 seconds, the defender closed in.

This is what tables do not tell you. Or they tell you, but not clearly enough. This is why I always keep a notebook beside the data table.

I extended the analysis to PPDA to test another hypothesis. Was Team C losing because their pressing was ineffective? Team C's average PPDA this season is 9.7 — meaning opponents complete only 9.7 passes before Team C takes a defensive action. That is the fourth-best figure in the league. They press well. They win the ball in the opponent's half 8.3 times per match — second-best in the league. In other words, they are playing the right way. They simply are not scoring.

There is a type of team that the table always treats unfairly: the team that does everything right except convert advantage into points. This season, Team C is that team.

Team D is the reverse. Team D sits fourth in the table, ninth in xG. PPDA 13.2 — below the league average. They play a low block, wait for the opponent to err, and exploit it well. Twelve of their 27 goals came from turnovers in the opponent's own half. This is a reasonable model in terms of results, but expensive in long-run probability. Across the last seven seasons, no club has sustained a top-four position with an xG ranking outside the top eight for two consecutive seasons.

Contrarian — Correlation Is Not Causation, and Where I Was Wrong

This is the hardest part of any data analysis. Because I have to name the thing my own model cannot explain.

In 2026-2026, my model got 41 of 248 matches wrong — an accuracy rate around 83.5%. That sounds good. But when I broke it down by match type, there was one anomaly. In derbies and in matches with crowds above 15,000, model accuracy fell to 71%. In other words, in big matches, xG became less reliable.

I spent three weeks testing this hypothesis. I compared predicted xG with actual xG for both teams across 38 derbies. The result: the favored team's actual xG typically fell 0.3 to 0.6 below expectation, while the underdog's xG typically exceeded expectation by 0.4 to 0.8. This pattern repeated in the Ha Noi derby, the central derby, and the big southern clashes.

The easiest explanation is psychology. The more accurate explanation lies in match structure: in big matches, the weaker side tends to abandon ball control and instead defend and counter with discipline, reducing the quality of the stronger side's chances. Meanwhile, the stronger side, under pressure to win, pushes its line higher and exposes gaps it would not otherwise expose. This is not psychology. It is a measurable tactical consequence: the stronger side in derbies averages 6.2 players in the opponent's half at the moment of turnover, versus 4.1 in ordinary matches. The space behind the midfield line widens by 40%.

This leads to a consequence I had missed for years: my context coefficient adjusted for empty stands, weather, and travel distance — but never for match importance. That was a serious omission. From this season I have added a "big match" variable to the model, and my derby predictions have improved recently.

But there is one warning I want on the table. Adding a variable can make me believe I understand. That is the biggest trap in this profession. The more variables, the better the model fits the past, and the more likely it breaks in the future. Eight years ago at Hang Day, I learned that what a model cannot explain is also part of the data. I write down my error, not to show off, but to remember.

There is one more thing I must be honest about. Team C — the high-xG, low-goal side — may not have a problem. They may simply be going through a random streak that will self-correct over time. But they may also be producing an illusion of a good team. The difference between those two possibilities lies here: if they play the right way over the next 10 matches and still do not score, that is a tactical issue. If they score again, it was randomness. And I cannot know in advance. No one can. The sweet bet does not exist; there is only probability mispriced and sold correctly.

Takeaway — Signals for the Next Round

I do not predict the future. I only read ahead the way the past continues to operate.

With 248 matches of data, I look at the next five V-League rounds and log three signals.

First, Team A — with a positive differential of 7.6 goals — will face two of the five best defensive sides in the league. The probability that their positive differential narrows to below three goals after five rounds is 68%, based on a seven-season historical sample. I am not saying they will lose. I am saying the 7.6 will not stand still.

Second, Team C — high xG, few goals — will meet three opponents with below-average defenses. If they still fail to score four goals across those five matches, I will log it as a structural issue, not randomness. If they score five or more, my model needs adjustment. I wait to re-fit.

Third, the "big match" variable I added needs at least 20 more matches to reach a sufficient sample. I will draw no conclusion about it until the data is sufficient. This is the discipline I set for myself after the Hang Day shock: do not trust a model merely because it was just right.

After every match, I still sit alone in the stadium when the crowd has left. Footsteps on the concrete stand, the sound of cleanup, a security guard humming softly. In that silence, I calculate nothing. The model has finished running. I only listen.

And in that silence, I understand something that 248 matches of data cannot tell me: behind every number there is a person. A striker who hits the post three times in a season, a keeper who saves a shot my model rated 0.79 xG, a player who runs 11.4 km in a match his team loses 0-3. Data records their actions, but cannot record how they sit in the dressing room afterwards.

The crowd leaves, the model breaks, and I learn to hear the breath of an empty stand.

A Few Lines on Method, for Readers Who Want to Verify

If you want to verify the numbers above yourself, here is what I make public. In 2026-2026, I tracked 248 matches across the V-League and national cup. Every shot is logged with seven variables: pitch coordinates, shot angle in degrees, type of preceding pass (open play, corner, free kick, counter), number of defenders within three meters of the shooter, match state (leading, drawing, trailing, goal difference), time within the match, and foot used.

My xG model uses logistic regression, trained on 4,812 shots from the 2026 to 2026 seasons in the V-League and national cup. I do not use foreign models because the league's characteristics differ: higher defensive density, more variable pitch quality, and match tempo roughly 12% slower than other Southeast Asian leagues.

The context coefficient is applied in four layers: the crowd layer (over 10,000 fans, 3,000 to 10,000, under 3,000, empty stadium), the weather layer (heavy rain, light rain, dry, humidity above 85%), the travel layer (away travel under 300 km, 300 to 800 km, over 800 km), and the congestion layer (three matches in 10 days, two in seven days, or a rest of more than seven days).

From 2026-2026, I added a fifth layer: match importance, categorized as derby, title decider, relegation decider, and ordinary match.

No model is right forever. The only thing I guarantee is that every number I publish is traceable. If you find an error in this piece, send it to me. I will add it to the dataset.

---

This analysis draws on public data and my own notes from the 2026-2026 season. All player and goal figures are recorded at the moment of observation. The piece offers no betting recommendation of any kind. Football carries high uncertainty; all conclusions are probabilistic, not assertions.