Regression to the Mean: Why Hot Streaks Do Not Last in Betting

A team wins three matches in a row and their odds to win the next match shorten significantly. A team loses three in a row and their odds lengthen just as dramatically. The market is responding to recent results. But recent results are not always reliable indicators of true quality. Teams on hot streaks tend to cool off. Teams on bad streaks tend to improve. This is regression to the mean, and it's one of the simplest but most profitable concepts in value betting.

Understanding Regression to the Mean: The Core Concept

Regression to the mean is a statistical principle that states: when a variable is extreme in one measurement, it will tend to be closer to the average in a subsequent measurement. This is true across almost every domain. Students who score very highly on one test tend to score closer to their average on the next test. Workers who have exceptional productivity in one quarter tend to regress towards their typical productivity levels in the next quarter.

In football, a team's true quality can be measured by their underlying metrics: Expected Goals, passing accuracy, shot conversion rate, etc. Let's say a team's true quality level suggests they should win 52% of their matches. Over a five-match sample, they might win four matches (80% win rate) simply by getting lucky. Over a 20-match sample, luck evens out and they'll win closer to 52%. Over a full season, the true quality determines the outcome.

The market prices matches partly on results and partly on underlying quality. When a team goes on a hot streak, the market overweights the results and underweights underlying quality. This creates a pricing inefficiency. The team's odds become too short because the market thinks they're better than they are. Conversely, teams on bad streaks get overlong odds because the market thinks they're worse than they are.

How Regression Applies to Football Results

Consider a real scenario. A team is on a four-match winning streak. They won 3-2, 2-1, 2-0, and 1-0. Their odds to win the next match are 1.60 (implied 62.5% probability). But let's look at their underlying performance:

Match 1: Created 1.2 xG, conceded 1.8 xG. Won 3-2 (lucky). Match 2: Created 1.4 xG, conceded 1.5 xG. Won 2-1 (slightly lucky). Match 3: Created 1.6 xG, conceded 1.2 xG. Won 2-0 (deserved). Match 4: Created 1.0 xG, conceded 1.4 xG. Won 1-0 (lucky).

Average: 1.3 xG created, 1.4 xG conceded. In terms of underlying performance, this team is actually slightly negative (conceding more than creating). Their four-match hot streak masks weaker underlying quality.

Based on this underlying data, their true win probability is probably closer to 48%, not 62.5%. At 1.60, betting against them offers value. Not because you think they'll definitely lose, but because the odds are overpricing their probability. Over multiple such bets, this edge compounds.

The key insight: the more extreme the recent results relative to underlying metrics, the more regression is likely. A team on a four-match winning streak with solid underlying metrics (creating more xG, defending well) is less likely to regress than a team on a four-match winning streak despite poor underlying metrics.

Why the Market Overreacts to Recent Form

Human psychology explains this. Bookmakers price matches using both algorithms and human traders. Retail bettors (the majority of their customer base) overwhelmingly overweight recent results. A team that's won three in a row gets backed heavily. This creates imbalanced liability. The bookmaker shortens the odds on that team to encourage bets on the opposition and balance their book.

Additionally, news coverage amplifies the recency bias. A team on a hot streak gets positive media coverage, which increases bet volume on them. A team on a bad streak gets negative coverage and negative bet flow. The bookmaker, managing real-time liability, adjusts odds towards these money flows. This sometimes moves odds away from true probability.

Professional bettors and sophisticated models don't overweight form as much. They look at underlying metrics. This creates a gap between what casual bettors think (a hot team is very likely to win) and what the data suggests (the team's underlying quality is average). The bookmaker's odds reflect somewhere in between. Value bettors exploit this gap.

Identifying Regression Opportunities

Here's how to spot teams likely to regress:

Step 1: Identify recent form extremes. Find teams with either very strong recent results (4+ wins in last 5) or very poor recent results (4+ losses in last 5).

Step 2: Gather underlying metrics. Check their xG created and conceded over the recent period. If the team is winning but underperforming xG (creating less than they're conceding, or getting lucky with conversion), regression is likely.

Step 3: Check season-long trends. Is this a hot streak the team is on, or a reversion to their actual pattern? A team that's been mediocre all season suddenly winning three matches in a row is more likely to regress than a team that's been strong all season.

Step 4: Estimate true probability. Based on underlying metrics and season-long performance, what do you think the team's actual win probability is? Is it significantly different from the bookmaker's odds?

Step 5: Look for mispricing. If the market has priced the team much shorter than your estimate suggests, consider betting against them (or betting the draw or opposition). If the market has priced them much longer, consider backing them.

Step 6: Track results. Over many such bets, you should see that backing teams likely to regress upward (bad streaks but good underlying metrics) and betting against teams likely to regress downward (hot streaks but poor underlying metrics) is profitable.

Real World Examples: When Regression Worked

Example 1: The Fortunate Winner. A mid-table team wins three matches in a row 1-0, 2-1, 1-0. Their odds to beat a strong team in match four drop from 3.50 to 2.80. But their underlying data shows they've been creating 0.9 xG per match while conceding 1.3 xG. They've been lucky to win. The strong team, despite losing their last match, has consistently created 1.6 xG while conceding 0.8 xG.

Backing the strong team at 1.45 (implied 69% probability) when your data suggests they have 70%+ probability is value. The market overweighted the three-win streak of the weaker team and underweighted their underlying quality gap.

Example 2: The Undervalued Faller. A top-four team loses two matches, both while creating 1.8 xG but getting unlucky in finishing. Their odds to beat a mid-table opponent drop from 1.40 to 1.65 after these two losses. The market has panicked. But the underlying data shows the team is creating chances at an elite rate and conceding very little. Backing them at 1.65 when 1.40 seems more appropriate is value.

In both cases, the regression principle tells you where the market has overreacted to recent results.

The Dangers of Over-Applying Regression

Regression to the mean is powerful but not absolute. Some teams genuinely improve or decline. A new manager might come in and change the team's underlying quality. A key injury might hurt performance. Tactical adjustments might make the team better or worse.

Also, regression works better over larger samples. Over 30 matches, regression to true underlying quality is nearly guaranteed. Over 5 matches, regression is likely but not certain. Always combine regression analysis with other contextual factors.

Additionally, you need to be careful about what "true quality" actually is. Using only two or three matches of data to define a team's underlying quality is dangerous. A team needs a larger body of work (10+ matches) before you can confidently say they're creating too many or too few chances.

In Summary

• Regression to the mean is the statistical tendency for extreme results to revert towards average performance
• The market overweights recent results relative to underlying metrics like Expected Goals, creating mispricings
• Teams on hot streaks are often too short; teams on cold streaks are often too long, creating value opportunities
• Identify teams whose recent results diverge significantly from underlying performance (xG, shot quality, possession)
• Value exists betting against hot teams (expect regression down) and for cold teams (expect regression up)
• The process: identify recent form extremes, gather underlying metrics, check season-long trends, estimate true probability
• Regression works better over larger samples; a two-week hot streak is less reliable than a two-month hot streak
• Regression is most powerful when combined with contextual knowledge about injuries, tactical changes, or management shifts
• At least 10 matches are needed for a reliable picture of underlying quality; fewer than 5 matches is insufficient data

FAQ

Q: How many matches should I look at to assess underlying quality? A: At least 10 matches for a reliable picture. 5 matches is a rough guide. Below 5 matches, you don't have enough data to distinguish true quality from luck.

Q: Can regression to the mean predict the exact next match result? A: No. Regression to the mean tells you that a team on a hot streak is likely to perform closer to their true level in the next match. But "closer to their true level" might still be a win. It predicts general direction, not specific outcomes.

Q: What if a team's hot streak is due to playing weaker opponents? A: This is good context. If a team won four in a row against bottom-half opponents but are now playing a top-four side, they're even more likely to regress. Their underlying quality didn't improve, just their recent opposition level.

Q: Is there a limit to how much regression I should expect? A: Regression happens gradually, not overnight. A team on an extreme hot streak might regress over 5-10 matches, not in the next single match. This is why tracking closing line value over large samples matters.

Q: How do I know if a team's underlying quality has actually improved vs. just been lucky? A: Monitor underlying metrics over time. If their xG creation is consistently improving match-to-match, something has changed tactically or in form. If their xG creation has stayed flat but results improved, it's luck.

Q: Should I ignore recent form completely and only look at underlying metrics? A: No. Recent form provides some information, especially recent underlying metrics. A team that's weak in underlying metrics and results is probably genuinely poor. A team with excellent underlying metrics but poor results is a value play. Use both.