A data-driven look at how score stability after the 87th minute connects to modern prediction markets — and what 79.3% actually means for traders
Tags: #soccer #data-analysis #sports-betting #statistics
There's a moment in nearly every close soccer match — somewhere around the 87th or 88th minute — when the crowd seems to collectively exhale. The team protecting a lead has run the clock down to almost nothing. The trailing team's chances of scoring feel increasingly theoretical. Broadcasters start wrapping up their analysis. Casual fans reach for their phones.
But what does the data actually say about that moment?
Is the late-game score really as stable as our intuitions suggest? And more importantly: does the prediction market ecosystem — platforms like Polymarket, where real money rides on real-time outcomes — accurately price that stability?
To answer those questions, we dug into 1,085 professional soccer matches drawn from StatsBomb's open dataset, spanning 41 competitions across multiple leagues, seasons, and continents. What we found was a number that has become something of a benchmark in our research: 79.3%.
That's the rate at which a score observed at the 87th minute held through the final whistle. Not a guess. Not a model extrapolation. A straight empirical count across over a thousand professional matches.
This article is about what that number means — for sports analytics, for prediction markets, and for anyone trying to think more clearly about late-game probability.
Why the 87th Minute? The Logic Behind the Cutoff
Before we get into the findings, it's worth explaining why we anchored our analysis at minute 87 rather than, say, minute 80 or minute 90.
The answer is practical and somewhat counterintuitive. Minute 90 is nominally the "end" of regulation, but soccer's stoppage time rules mean the game effectively continues for anywhere from one to ten or more additional minutes. Choosing minute 90 as a cutoff would systematically exclude a meaningful chunk of late-game action — including many of the most dramatic goals in the sport's history.
Minute 87 is different. It's early enough to capture matches across a wide range of typical stoppage time allowances (2–5 minutes in most standard matches), but late enough to genuinely test what "late-game stability" looks like in practice. It's also a timestamp that aligns reasonably well with how in-play prediction markets behave — liquidity tends to thin out, odds compress, and traders implicitly treat the current score as increasingly durable.
Think of minute 87 as the moment when prediction markets are essentially pricing a very short-term binary: does this score hold, or doesn't it?
Our dataset gave us 1,085 instances of that exact question, with definitive answers recorded at full time.
The Core Finding: 79.3% and What It Actually Tells Us
Across all 1,085 matches in our sample, the score at minute 87 remained unchanged at full time in 860 cases — a rate of 79.3%.
Let's sit with that number for a moment before unpacking it.
79.3% sounds high. And in absolute terms, it is. If you were asked to bet on whether the score would change in the final three-ish minutes of a professional soccer match, knowing nothing else, the historical base rate says you'd be right about 4 times out of 5 if you bet "no change."
But 79.3% also means that roughly 1 in 5 matches saw a score change after minute 87. That's not a trivial tail risk. In a world where prediction market prices sometimes approach 90–95% for a "leading" outcome in the final minutes, that historical base rate provides an important reality check.
Prediction markets are supposed to aggregate information efficiently. If the market is pricing a current leader at 93% with three minutes left, but the empirical base rate for score stability is 79.3% across similar situations, something in that pricing warrants scrutiny — either the market knows something the base rate doesn't (game state, fatigue, red cards, pressure), or it's systematically overconfident in the final minutes.
The answer, unsurprisingly, is somewhere in between — and the score-level breakdown helps explain where.
Score-by-Score Breakdown: Not All Leads Are Created Equal
One of the most important findings in our analysis is that the 79.3% aggregate conceals significant variation by score state. This is exactly the kind of granular data that sophisticated prediction market participants should be tracking — and that many currently aren't.
Here's the breakdown for the four most common score states observed at minute 87 in our dataset:
0-0 Draws: 82.3% Stability
Scoreless matches at minute 87 held through full time at a rate of 82.3% — actually higher than the overall average.
This might seem counterintuitive. Surely a 0-0 game is more likely to end 0-0 than a 1-0 game is to end 1-0? Not necessarily — but the data suggests that 0-0 games in the late stages often reflect two evenly matched, defensively oriented teams who have spent 87 minutes failing to score. The marginal probability of late scoring actually decreases in such contexts.
For prediction market traders, this is meaningful. A 0-0 draw market at minute 87 should theoretically price "draw" at something close to 82%, not dramatically higher or lower.
1-0 Leads: 79.7% Stability
The classic one-goal lead is the most psychologically interesting case. It's the scenario that spawns the most late-game anxiety for fans and the most active trading on in-play markets.
At 79.7%, one-goal leads at minute 87 are highly — but not overwhelmingly — likely to hold. The ~20% failure rate is real and meaningful. It accounts for the full range of late equalisers: set-piece goals, individual brilliance, goalkeeping errors, desperate headers in stoppage time.
If a prediction market is pricing a 1-0 leading team at, say, 88% to win with three minutes left, our data suggests the market is somewhat overvaluing that lead relative to pure base rates. The correct baseline for 1-0 stability is closer to 79.7%.
0-1 Deficits: 79.0% Stability
The mirror image of the 1-0 lead, at 79.0% — very close to the 1-0 case, as you'd expect. The slight asymmetry (0.7 percentage points lower stability than 1-0 leads) is interesting but not statistically dramatic. It potentially reflects a marginal tendency for trailing home teams to press harder in the final minutes, creating slightly more chaotic game states that occasionally produce equalisers.
1-1 Draws: 76.6% Stability
This is the most important outlier in the dataset, and the one that most directly challenges prediction market efficiency.
76.6% of 1-1 score states at minute 87 held through full time. That means nearly 1 in 4 drawn matches saw a late change — a substantially higher instability rate than any other score state we examined.
Why? Several factors likely contribute:
Mutual motivation asymmetry: In a 1-1 game, both teams have some incentive to win — but also a clear floor outcome (the draw). Neither team is fully committed to pure defense in the way a 1-0 leader might be.
Open game dynamics: Matches that reach 1-1 have typically seen attack-oriented periods from both teams. That attacking pattern often continues late into games, creating more dangerous situations.
Set piece exposure: Both teams have had at least one goal, suggesting both defenses have shown some vulnerability. Late corners, free kicks, and throw-ins create genuine opportunities.
Physical fatigue affecting defensive organization: Teams that have traded goals through 87 minutes have expended significant energy on both scoring and recovery.
For prediction market traders specifically, the 1-1 score state at minute 87 is perhaps the most mispriced situation in late-game soccer markets. If you're seeing "draw" priced at 85%+ in a 1-1 game with three minutes left, the historical base rate says 76.6%. That gap is worth understanding.
What This Means for Prediction Market Efficiency
Modern prediction markets — Polymarket being the most prominent example for sports and event outcomes — have made enormous strides in pricing accuracy. Academic research consistently shows that well-functioning prediction markets outperform polling averages, expert panels, and simple statistical models on many types of events.
But late-game soccer is a particular kind of challenge for these markets.
The core problem is liquidity and participation asymmetry in the final minutes. As a match enters the 87th, 88th, 89th minute, several things happen simultaneously:
- Casual participants exit: Many traders close positions or simply stop participating as the outcome feels "decided"
- Bid-ask spreads widen: Reduced liquidity means market makers demand larger spreads to compensate for uncertainty
- Automated bots may anchor to recent ticks: Algorithmic participants can lag real-world events by crucial seconds
- Emotional overweighting of visual salience: A team that looks dominant visually may be overpriced relative to actual goal probability
These dynamics create predictable inefficiencies — not necessarily large or consistently exploitable ones, but real patterns that show up in the data.
The 79.3% benchmark gives prediction market analysts a concrete reference point. When a market implies a winning probability above 90% for the current score to hold at minute 87, it's asserting something that the historical base rate doesn't support. When it prices below 75%, it's potentially overcorrecting for volatility that rarely materializes.
The most useful application isn't necessarily direct trading — it's calibration. Good prediction market participants, like good forecasters of any kind, should be able to compare their model probabilities against base rates and understand when they're diverging and why.
