How reliable is Poisson for predicting corners?

Poisson provides a useful baseline but has limitations. It assumes constant average rate, which corners violate due to game state and style. For short-term predictions, team-specific averages add some accuracy, but model error remains high.

What is a good sample size for team corner averages?

At least 20-30 matches to reduce noise. Corners have higher variance than goals, so smaller samples can mislead. Use recent matches (last season if squad stable) and adjust for opponent strength.

Should I use league average or team average for λ?

Start with league average as a baseline, then adjust for team-specific rates. For the total corners market, use the sum of both teams' expected corners. Blended estimates often beat pure league averages.

Can the Poisson model be used live?

Pre-match only, because game state changes the rate. Live models require real-time updates (like in-play shot and possession data) which are beyond simple Poisson. For live betting, look at momentum indicators instead.

Corner Kicks: A Quantitative Approach to Betting on Set Pieces

Updated on 2026-07-27 · 713 words

What it is

Corner kick markets offer bets on the total number of corners in a match (over/under a line, exact count, or team totals). Unlike goals, corners are more frequent but also more volatile. They are influenced by playing style, game state, and opposition, making them a distinct market where quantitative analysis can add clarity. The Poisson distribution is commonly used to model the number of corners, treating each match as an independent event with a fixed average rate.

The math

The Poisson distribution gives the probability of exactly k events occurring when the average rate is λ:

P(X = k) = (λ^k × e^(-λ)) / k!

To price an over/under line, we need the cumulative probability up to the line. For a line of 9.5 corners (over means 10 or more), we calculate P(over 9.5) = 1 - P(X ≤ 9). P(X ≤ 9) is the sum of P(X = 0) through P(X = 9).

For example, with λ = 10.5 (average total corners per match), we compute:

P(X ≤ 9) ≈ 0.392 (using tables or software)
So P(over 9.5) = 1 - 0.392 = 0.608

Implied probability from American odds: if the line is priced at -110 (decimal 1.909), the implied probability is (110 / (110+100)) = 0.5238, or 52.38%. The market is assigning a 52.38% chance to over 9.5, while the Poisson estimate gives 60.8%. The difference suggests value if the estimate is accurate.

But markets have margins. The sum of over and under probabilities will exceed 100%. To find the fair probability, we remove the margin proportionally. If the over is -110 (52.38%) and under is -110 (52.38%), total = 104.76%, margin = 4.76%. Fair over probability = 52.38% / 104.76% = 50.0%. That is the breakeven point.

Worked example

Team A vs Team B: historical average total corners is 10.2. we want to bet over 9.5. The bookmaker offers over 9.5 at -120 (decimal 1.833). Implied probability: 120 / (120+100) = 0.5455 (54.55%).

Using λ = 10.2, we calculate P(over 9.5) = 1 - P(X ≤ 9). P(X ≤ 9) for λ=10.2 is approximately 0.380 (from Poisson cumulative). So P(over 9.5) = 0.620 (62.0%).

Compare: 62.0% (estimate) vs. 54.55% (market). The estimate is higher. To find expected value: EV = (0.620 * 1.833) - 1 = (1.136) - 1 = +0.136, or +13.6%.

Even after adjusting for margin (the over/under market might have a 5% margin, so fair probability ≈ 54.55% / 1.05 = 51.95%), the estimate remains above. This is a candidate for a positive expected value bet.

However, note that the Poisson model is a simplification. Actual corner rates depend on real-time conditions, team tactics, and match events.

When not to use it

Game state heavily changes corner rates. Teams losing often push forward, forcing more corners. Poisson models using pre-match averages ignore the impact of scoreline, which can shift λ by 20-30% within a match. In-play adjustments are necessary.
Style of play dominates the average. A team that plays direct, crossing football generates more corners than a possession-based team, even if overall quality is similar. Using league-average λ hides these differences; team-specific averages (with sufficient sample) are essential.
Corner data is less reliable than goal data. Corners are rarer than shots but more frequent than goals. However, the sample size available for corner averages is often smaller, especially for lower leagues or after squad changes, increasing estimation error.
Market efficiency is lower for corners. Corner markets attract less liquidity and sharper action than goal markets. Margins can be higher (up to 10% in some books), and odds may not reflect all available information. The Poisson advantage can be nullified by inflated juice.
No model captures in-play adjustments. Corners are influenced by red cards, tactical switches, and set-piece efficiency. A static Poisson model cannot account for these dynamic changes, limiting its usefulness to pre-match analysis only.

For all these reasons, treat corner betting as a high-variance market where even a positive EV estimate requires a large edge and disciplined bankroll management.

Use the analyzer tool to compute Poisson probabilities and compare with market odds.

21+ only. Gambling involves financial risk of loss. Please wager responsibly. If gambling ceases to be entertainment, seek help.

Frequently asked questions

How reliable is Poisson for predicting corners?: Poisson provides a useful baseline but has limitations. It assumes constant average rate, which corners violate due to game state and style. For short-term predictions, team-specific averages add some accuracy, but model error remains high.
What is a good sample size for team corner averages?: At least 20-30 matches to reduce noise. Corners have higher variance than goals, so smaller samples can mislead. Use recent matches (last season if squad stable) and adjust for opponent strength.
Should I use league average or team average for λ?: Start with league average as a baseline, then adjust for team-specific rates. For the total corners market, use the sum of both teams' expected corners. Blended estimates often beat pure league averages.
Can the Poisson model be used live?: Pre-match only, because game state changes the rate. Live models require real-time updates (like in-play shot and possession data) which are beyond simple Poisson. For live betting, look at momentum indicators instead.