Glossary

ROI (Return on Investment)

Return on investment. It shows how much you won or lost overall per 100 units staked. E.g. +12% ROI means 100 became 112.

ROI = (total winnings − total stake) / total stake × 100

EV (Expected Value)

Shows how much a bet is statistically „worth”. With positive EV the bet's mathematical expected value would indicate profit over the long run, but only if the probability estimate is accurate; in practice that estimate is the hard part.

EV = (probability × odds) − 1. If EV > 0, the bet is „valuable”.

Hit Rate

The share of correct predictions among all forecasts. A 60% hit rate means 60 of 100 predictions were correct.

Odds

The multiplier offered by the bookmaker. In decimal format: 2.00 means 100 staked returns 200 (100 winnings + 100 stake).

At odds of 1.85: 100 → 185 returned, 85 profit.

Brier Score

A measure of the quality of a probability forecast. It ranges from 0 to 1; the smaller, the better the model. 0 for a perfect model, 1 for a completely wrong one.

Calibration

Shows how well the model's probabilities match real frequencies. If the model gives a 60% probability for a home win, it should actually occur about 60% of the time.

CLV (Closing Line Value)

The difference between the pre-match closing odds and the odds available when you bet. Positive CLV shows you bet at a better price than the one the market finally calibrated to.

Poisson distribution

A statistical distribution that models the count of rare, mutually independent events. In football it is used to estimate the number of goals.

If a team scores 1.5 goals per match on average, the Poisson distribution gives the probability of scoring 0, 1, 2, 3, etc. goals.

Elo rating

A ranking system tracking team strength, originally developed for chess by Arpad Elo. The winner's and loser's scores update after every match.

Monte Carlo simulation

A method based on random simulations. It samples the match's goal counts from both teams' expected goals (Poisson distribution) many thousands of times and computes the probabilities from the frequency of the simulated results.

Accumulator

Combining several matches into a single bet. The odds multiply and the payout can be large, but every match must come in.

System bet

A system of combinations where not every match has to come in. A smaller payout but more safety.

Margin (bookmaker's cut)

The bookmaker's built-in profit in the odds. E.g. if a match's true probability is 50-50 but the bookmaker offers 1.90-1.90 (not 2.00-2.00), the margin is ~5.3%.

Value betting

A bet where the model thinks the true probability is higher than the odds reflect. In theory this is positive EV; in practice it is only worth something if the model is more accurate than the market, which our research shows is rare.