What Is Correlation?
Correlation is when two events are not independent. The outcome of one affects the probability of the other.
Example:
- Weather affects multiple matches simultaneously
- Team form affects performance across multiple fixtures
- Injuries to same team affect multiple legs in your acca
When your acca legs are correlated, your actual win probability is lower than calculated odds suggest.
Understanding Acca Correlation
Uncorrelated legs (independent):
- Manchester City to win + Liverpool to win
- These teams play different opponents at different times. One team's result doesn't affect the other.
- Actual probability ≈ 0.66 × 0.66 = 43.6%
Correlated legs (dependent):
- Manchester City to win + Over 2.5 goals (same match)
- These outcomes are linked. If City wins 2-1, over is more likely than 1-0.
- Actual probability ≠ simple multiplication
- Actual probability < naive calculation
Extreme correlation:
- Manchester City to win + Manchester City to score first
- City can't score first and lose. These are highly correlated.
- Actual probability significantly lower than naive calculation.
Common Sources of Acca Correlation
Same team, multiple fixtures:
- Leg 1: Brighton to win
- Leg 2: Brighton to score 2+
- Leg 3: Brighton to win with BTTS
All depend on Brighton's performance. If Brighton is off form, all three might fail together.
Same match, multiple markets:
- Leg 1: Home win
- Leg 2: Over 2.5 goals (same match)
Home winning is more likely with high scoring. These are correlated.
Weather conditions:
- Three matches scheduled for same day, heavy rain forecast
- All three matches might be affected similarly
- Over/under bets in all three correlated downward
Fixture congestion:
- Team playing midweek European match + weekend domestic match
- Team might be fatigued and rotation at weekend
- Correlation between fixtures.
League-wide trends:
- Building acca on all top-six teams in one weekend
- League-wide form trends affect all of them similarly
- Correlation across all legs.
Impact of Correlation on Odds
Uncorrelated three-leg acca (independent):
- Odds: 1.80 × 1.80 × 1.80 = 5.832
- Calculated probability: 55.6%^3 = 17.2%
- Actual probability ≈ 17.2% (independence assumed correctly)
Correlated three-leg acca (same match, multiple markets):
- Odds: 1.80 × 1.80 × 1.80 = 5.832 (same odds)
- Calculated probability: 17.2% (naive calculation)
- Actual probability: ~12-14% (correlation reduces real probability)
You're getting 5.832 odds but real probability is 13%, not 17.2%. The odds are worse value than they appear.
Identifying Correlations
Red flags for correlation:
-
Same team, multiple legs: Any acca with two legs involving the same team is correlated.
-
Same match, multiple markets: Home win + Over 2.5 goals is correlated.
-
Multiple matches affecting same team: Team plays Wednesday and Saturday. Your acca has both. Fatigue correlates them.
-
Geographic clustering: Three matches all in cold weather, or all in wet conditions.
-
Form dependencies: All legs depend on similar team form patterns.
Calculating Correlation Impact
Pearson correlation coefficient (0 to 1):
- 0 = no correlation (independent)
- 0.5 = moderate correlation
- 1.0 = perfect correlation (same outcome)
Without statistical tools, estimate:
- Home win + Over 2.5 (same match): 0.3-0.4 correlation
- Home win + Home team to score 2+: 0.5-0.6 correlation
- Same team in two matches: 0.4-0.5 correlation
Higher correlation = lower actual probability than calculation suggests.
Correcting for Correlation
Simple adjustment:
If you calculate 17.2% probability but suspect 40% correlation:
- Reduce calculated probability by correlation factor
- 17.2% × (1 minus 0.4) = 17.2% × 0.6 = 10.3% actual probability
This is crude but better than ignoring correlation.
Avoid rather than calculate:
Easier approach: simply avoid correlations.
- Don't combine same team with multiple markets
- Don't combine same match legs
- Don't combine same-day matches dependent on same conditions
Building Uncorrelated Accas
Best approach:
- Different teams across different matches
- Independent market combinations (match result, not home win + same-match over)
- Different conditions or days
Example:
- Leg 1: Man City to win (Friday match)
- Leg 2: Liverpool BTTS (Saturday match, poor weather)
- Leg 3: Chelsea to win (Sunday match)
These are largely independent. Different teams, different days, different conditions.
When Correlation Is Unavoidable
Sometimes correlation is inherent to your analysis:
- You believe "both teams in this match will attack" (home win + BTTS likely together)
- Correlation is real, but it matches your analysis
In this case:
- Acknowledge correlation explicitly
- Discount your calculated probability by estimated correlation factor
- Accept that real probability is lower than odds suggest
- Only build acca if you're still confident
Correlation in Same-Game Multis
Same-game multis are inherently correlated (all from one match). Legs within one match are more correlated than legs across matches.
This is actually beneficial if understood correctly:
- Real probability is lower than calculated
- But odds often don't fully price in this reality
- You might find value by using SGMs instead of multi-match accas
Testing for Correlation
Over time, tracking reveals correlation:
-
Build accas with same-team correlation
-
Track actual hit rates
-
Compare to calculated probabilities
-
If actual is consistently lower, correlation was real
-
Correlation means acca legs are not independent.
-
Common sources include same team in multiple legs, same match multiple markets, shared weather conditions, and league-wide trends.
-
Correlation reduces actual probability below calculated odds suggest.
-
Same-match legs are 30-60% correlated.
-
Same-team legs are 40-50% correlated.
-
Same-day weather conditions create 20-40% correlation.
-
Identify correlations and either avoid them (build truly independent accas) or adjust probability estimates downward.
-
Test correlation effects through tracking over time.