The Transfer Window, Small Samples, and the Discipline of Saying: Not Enough Data

July in Da Nang, the rain comes late

I was sitting on the second floor of a coffee shop on Nguyen Van Linh Street, a long-cold cup in front of me and a spreadsheet open on the screen. The spreadsheet held three rows of data about a striker whom three separate social media accounts had just declared was on his way to a V.League club. Three rows. One name. A transfer fee nobody had confirmed. And a gap larger than everything I actually had.

I sat there for two hours. The only conclusion I managed to write down was: not enough data to conclude.

That answer is not interesting. But it is correct. And across seven years of watching the sports market through a data lens, I have learned that most analytical errors do not come from misreading a number. They come from daring to conclude when the sample size has not yet granted permission.

Context: the economy of rumour

The transfer window is the strangest stretch of the year for anyone whose job is analysis. During a season, we have a relatively stable frame of reference: fixtures are played on schedule, line-ups are published, match data is logged minute by minute. Everything is verifiable. Then June and July arrive, and that entire system disappears.

What replaces it is an information market with no exchange, no regulator, and no disclosure obligation. An account with two hundred followers can generate a wave of opinion if the content is compelling enough. An agent can drop a hint in an interview, and within six hours that hint becomes a headline across the press.

In Vietnam, this has its own specific features. The V.League has no centralised transfer disclosure system of the kind European leagues operate. Clubs announce what they choose to announce. Contracts rarely have release clauses that are made public. Domestic transfer fees are almost always kept confidential, or announced in round figures that no longer carry analytical meaning. Under those conditions, fans consume information from unofficial channels, and the media competes on speed rather than accuracy.

I do not see this only in football. Vietnam's esports scene runs on a similar mechanism. When the player transfer market opens, every piece of information about a player leaving a team originates from a livestream, a status update deleted within minutes, a clipped video. Any experienced analyst recognises it immediately: most of this is noise, not signal.

The problem is that the line between noise and signal is not always clear. Which is why we need a system.

Core: small samples are polite liars

In statistics there is a concept every sports analyst should carve into their desk: the law of large numbers. As the number of observations grows, the average value of a variable converges on its true value. When the number of observations is small, the average can sit very far from the true value — and, worse, it can sit there in a way that looks entirely convincing.

Imagine a striker who plays three games at the start of a season and scores four goals. That rate extrapolates to more than one hundred and thirty goals across a thirty-eight-game season. Nobody believes that. But if that same striker scores three goals in two pre-transfer-window friendlies, articles appear immediately declaring he has reached a new level.

The psychology here is straightforward. Humans look for patterns, even when the pattern is built on a sample too small to mean anything. Analysts are not immune. In fact, we are more exposed, because we have the tools to give an appearance of precision to conclusions that have no foundation.

That is why I begin every piece of analysis with a single question: what is the sample size here?

In transfer questions, the sample usually falls into one of three buckets. The first is the player's previous season, usually large enough if he played more than twenty matches. The second is national team data, where the sample is often four to ten matches a year. The third is data from friendlies and matches in a new league, where the sample may be two or three.

These buckets do not carry equal weight. Blending them without weighting is one of the most common errors I see in transfer reports.

PPDA is a lens — through it, I saw Morocco in the semi-finals two months early

In 2026 I was nineteen, in my first year of university. Before the World Cup in Qatar, I built a ranking of the thirty-two teams based on three years of defensive data: PPDA, average distance covered, and shots faced inside the box. The model placed Morocco in the top eight. All my friends laughed.

They reached the semi-finals.

The point is not the story of a correct prediction. The point is the structure of the test. That model was not built on one match. It was built on three years of continuous defensive data, with samples running into hundreds of matches per team. When the sample is large enough, signal rises out of noise. And Morocco, across those three years, was one of the most structurally stable defensive teams in the world, in a way the league table could not express.

Conversely, when the sample is small, anything can look like signal. A team that wins three straight friendlies is described as being in form. A player who scores twice in two games is described as hitting his stride. But three games and two games are not enough to separate ability from luck.

During the transfer window, almost all the information available to us is small-sample. That is the nature of this market, and denying it helps nobody.

Three layers of reliability filtering

Over the years I have built a three-layer system for assessing any transfer information before it enters analysis.

Layer one is provenance. Where did this come from? If it comes from an account with no track record, it should not be treated as a fact. If it comes from a player's agent, it needs to be read as part of a negotiation strategy, not as a statement of truth. If it comes from the club, it carries the highest weight — but even then, the timing of the announcement carries information of its own.

Layer two is cross-verification. Information only has value if it appears simultaneously from at least two independent sources. In the case of the player I was analysing in that coffee shop, three sources reported it, but all three led back to one original post. That is not three sources. That is one source copied three times.

Layer three is structural feasibility. Even if the information is true, is it possible? Can a club whose wage bill has already hit its ceiling sign a major contract? Can a player with two years left on his deal leave on a free? These questions do not require secret data. They require logic.

These three layers do not eliminate error. But they eliminate most of the information that, if fed into analysis, would corrupt every result downstream.

When the result is zero

There is something school does not teach us adequately: in research, a null result is a valid result.

If an experiment shows no effect, that does not mean the experiment failed. It means the hypothesis was not supported, and that is valuable information. But in sports analysis, null results are almost never published, because nobody wants to read a piece concluding that there is nothing to conclude.

This is a more serious problem than it appears. When only strong conclusions are published, readers form the impression that the sporting world can always be explained, that there is always a clear cause for every outcome. Reality is the opposite. Most of what happens on a pitch is the product of ability, luck, and variables nobody measures.

An honest analyst must be able to say: I do not know.

During the transfer window, that sentence appears more often than in any other period. And that is normal.

The case of the left wing and the right wing

In June 2026 I was interning at a small sports data company in Ho Chi Minh City. The European Championship was underway, and Spain fielded a teenage wide pair: one sixteen, one twenty-one. What caught my attention was not their age but the way they interacted with the team's structure.

I spent two weeks writing a twelve-page report I called the two-wing ecosystem. Its central finding: this wide pair generated more chances than any other pairing in the tournament, and most of it came from dribbles into central areas rather than crosses from wide. The right flank specifically was where the youngest player in the squad received the ball on average more than eleven times per match when opponents pushed up. Each such reception opened space for the full-back behind him to advance.

The important thing here is method. I did not start from the conclusion that the pair was good. I started by measuring receptions, reception locations, and the consequences of those receptions. The sample was six matches across the group stage and knockouts — enough to be meaningful, not enough to claim durability.

When my boss sent that report to three European betting companies, a firm in Malta sent a part-time job offer. I accepted it, but kept my studies. Because a six-match sample, however beautiful, is still a small sample. And I did not want to build a career on a small sample.

I do not watch football to enjoy it. I watch it to test a long-term hypothesis.

That sentence sounds cold, and it is cold. But it explains why I log every bet I place, along with the reasoning behind the decision, regardless of whether it wins or loses.

That log is not a tool for showing off results. It is a slow verification mechanism. Each entry is one observation. After a year I have a few hundred observations. After three years I have enough data to know which kinds of decisions I make well and which I make badly. No single winning bet can teach me that.

During the transfer window this log is especially useful. It forces me to write the reasoning before I know the outcome. If that reasoning is an unverified rumour, I see it immediately when I read back. And I know the decision was worth nothing, regardless of whether it turned out right.

Contrarian angle: the pressure to always have a take

In sports media there is an invisible but very strong pressure: the pressure to always have an opinion.

If you go quiet for a week, the audience leaves. If you say there is not enough data, you are seen as lacking conviction. If you refuse to predict, you are seen as lacking expertise. So many people in the profession choose to say something, even when that something is not true.

I understand the logic. I have felt it.

But there is a fact that people who must always have a take tend to overlook: every time you publish a rushed conclusion, you are not just betting on one match. You are betting on your own long-term credibility.

The sports analysis market runs on an accumulation mechanism. Someone who makes ten predictions and gets seven right will be trusted more than someone who makes five and gets all five right, if the second person always picks low-uncertainty situations. But if the first person makes twenty predictions and gets seven right, their rate has collapsed.

This is the crux: the number of conclusions you publish is not an asset. It is a liability if you have no basis for it.

During the transfer window that liability accumulates faster than at any other time. Every rumour you comment on is a small debt. By the end of August you may have accumulated hundreds of debts, with no way to repay them.

There is another fact worth stating: disagreeing for the sake of disagreement is not analysis. This is the trap people like me fall into most easily. We have gone against the crowd and been right, a few times, and that creates a confidence disproportionate to actual ability. From there, going against the crowd becomes a reflex, and a reflex is not analysis.

Before writing any contrarian view, I ask myself one question: is this right because the data shows it, or right because I want it to be?

Most of the time, the answer is the second one.

The limits of what can be predicted

There is a limit sports analysts are reluctant to admit: most match outcomes cannot be predicted from pre-match data.

This does not mean data is useless. It means data has a specific range of validity. Data can tell you which team creates more chances over the long run, which team has a more durable defensive structure, which player tends to improve steadily. Data cannot tell you who scores in the eighty-fourth minute.

During the transfer window that range narrows further. Whether a signing succeeds depends on variables data does not touch: cultural adaptation, dressing-room integration, the relationship with the coach, family stability. None of these appear in a spreadsheet.

An honest analyst draws that boundary clearly, rather than filling it with speculation presented as fact.

Takeaway: the signal arrives later, not now

After that spreadsheet with three rows and one large gap closed, I did not write about that player.

Two weeks later the transfer collapsed.

One could say I was right not to conclude. But I do not think of it that way. What I think about is all the articles written in the interval between those two moments, with the same amount of data I had, where the conclusion had already been decided in advance.

The transfer window will continue. There will be more rumours, more numbers, more credibility rankings shared around. Most of them will dissolve in silence, unmentioned, with nobody held accountable.

What I want to carry out of this season is not a correct prediction. It is a habit: look at the sample size before you look at the story. And if the sample is too small to answer, let the answer stay open. In football, as in every complex system, waiting for a signal strong enough to say something is not hesitation. It is the only way that when the signal arrives, it is still a signal.