Betting glossary: from xG to Asian handicap

A handicap that takes the draw out of the market entirely — it either refunds the stake or, on quarter lines, splits it across two halves. One side starts with a goal head start or deficit that is added to the final score. Because both prices sit close together, the margin is usually lower than on a plain 1X2 market.

Example: At −0.5 the home side simply has to win; at 0.0 a draw refunds the stake; at −0.25 a draw counts as half a loss.

The money set aside strictly for betting and kept apart from everyday funds. It is the denominator for stake sizing and the yardstick for risk — without a defined bankroll there is no way to say whether a stake is large or small. The basic rule is that any single bet is a small fraction of the whole.

Example: A €500 bankroll at a 1 % unit means €5 a bet, so a run of ten losses costs 10 % of the bankroll rather than all of it.

A bet on whether both sides find the net at least once. A 1-1 and a 3-2 are both BTTS yes; a 4-0 is BTTS no. The market ignores the winner completely and only asks whether goals arrive at both ends.

Example: Two attacking sides with high xG and high xGA make BTTS yes far likelier than a meeting of two low-block defences.

The price the market settles on just before kick-off. By then the line-ups, the weather, the absences and all the money wagered on the fixture are priced in. That is why the closing line is treated as the best publicly available estimate of an outcome’s probability.

Example: A price of 2.30 three days out drifts to 2.05 once the line-ups drop — the closing line is 2.05, not 2.30.

The gap between the price you took and the closing line. Positive CLV means you bet at better odds than where the market finished. It is the fastest feedback on selection quality, because it does not depend on whether that one bet happened to land.

Example: You take 2.20 and the closing line is 2.00. CLV is 2.20 / 2.00 − 1 = +10 %, even if the match went the other way.

A bet on the exact final score, say 2-1. There are dozens of possible scorelines, so each single one carries a small probability and a long price. That is also why bookmaker margin on this market tends to be among the highest on the board.

Example: If a model puts 1-1 at 12 %, the fair price is 1 / 0.12 ≈ 8.3 — and the offered price will sit below that once margin is applied.

A bet on the match winner where a draw returns the stake. You pay for that insurance with a shorter price than the equivalent 1X2 selection. Mechanically it is an Asian handicap of 0.0 under a friendlier name.

Example: A €20 DNB on the home side pays at the quoted price on a win, refunds €20 on a draw and loses on a defeat.

One bet covering two of the three results: 1X (home or draw), X2 (away or draw) or 12 (no draw). The higher hit rate is paid for with a much shorter price. Note that 12 is not the same as DNB — a draw loses the bet rather than refunding it.

Example: A 1X selection wins on both a 2-1 and a 1-1, and only loses if the away side wins.

A whole-number handicap with three outcomes: win, draw and loss once the head start is applied. Unlike the Asian version nothing is ever refunded — the handicap draw is its own selection you can back. It shows up mostly on fixtures with a heavy favourite.

Example: A home side at −1 wins the bet at 2-0; at 1-0 the handicap draw lands and the bet on the home side loses.

How a side has performed in its recent matches. On its own it is a weak signal, because a run of results says nothing about the opposition or the quality of the performances. It becomes useful in context — alongside xG, xGA and the strength of the schedule played.

Example: Five wins over bottom-of-the-table sides at 0.9 xG a game tell you less than three draws against the leaders.

The record of previous meetings between two sides. Only the slice where the competition, the venue and roughly the current squads still match carries information. Fixtures from several years ago under different managers are a historical footnote.

Example: A cup tie six seasons ago played with a reserve XI carries essentially no weight when previewing a league match.

The probability a price expresses, calculated as 1 divided by the decimal odds. It puts prices and model estimates on the same scale so they can be compared at all. It also contains the margin, so it always sits a little above the market’s genuine estimate.

Example: A price of 2.50 implies 1 / 2.50 = 40 %. A price of 1.40 implies 71.4 %.

A formula for how much of a bankroll to stake given a known edge and price. It maximises long-run growth but assumes the probability estimate is right — overstate the edge and it stakes far too much. That is why half Kelly or quarter Kelly is the common practical choice.

Example: At 2.20 with a 50 % estimate: (0.50 × 2.20 − 1) / (2.20 − 1) = 0.10 / 1.20 ≈ 8.3 % of bankroll. Half Kelly makes it 4.2 %.

The bookmaker's profit built directly into the prices. Odds are set so the implied probabilities of every outcome add up to more than 100 %, and that surplus is the margin. The lower the margin, the more of the fair price stays with the bettor.

Example: Prices of 2.00 / 2.00 on a two-way market give 50 % + 50 % = 100 %, so zero margin. Prices of 1.90 / 1.90 give 52.6 % + 52.6 % = 105.3 %, a margin of roughly 5 %.

The check of whether predicted probabilities match reality. A model is calibrated when situations it prices at 70 % come in roughly seven times out of ten. Calibration is not the same thing as accuracy — a model can be accurate and still overstate its own confidence.

Example: If only 55 of a hundred 70 % predictions land, the model is overconfident and its numbers need adjusting.

Testing a model on data it never saw while learning. Performance on training data always looks better than what the model achieves in practice, and the out-of-sample test is the guard against that. Without it there is no way to tell real predictive power from overfitting.

Example: A model trained on the 2018 – 2023 seasons and tested on 2024 gives a far more realistic picture than the same test run on the data it learned from.

A bet on the combined goal count against a stated line, most often 2.5. Over 2.5 needs three goals or more, Under 2.5 needs two or fewer. The half goal in the line is deliberate: it removes any chance of the total landing exactly on the number.

Example: A 2-1 gives three goals, so Over 2.5 wins and Under 2.5 loses. On a 3.0 line the same score would be a push and the stake comes back.

The sum of the implied probabilities across every outcome of a single market. Anything above 100 % shows how much the market is oversold. Margin follows from it as overround minus 100 %, or more precisely as 1 − 100 / overround.

Example: Prices of 2.10 / 3.50 / 3.60 give 47.6 % + 28.6 % + 27.8 % = 104.0 %. The overround is 104 % and the margin about 3.8 %.

A probability distribution for the count of independent events in a fixed interval — in football, for goals. Given a side’s expected goal count it returns the probability of 0, 1, 2 or more goals, and from there a full correct-score matrix. It is a simplification: goals are not perfectly independent, so practical models adjust it.

Example: At 1.5 expected goals the chance of a blank is e^(−1.5) ≈ 22 %.

Net profit divided by the total amount staked, shown as a percentage. It says how much each euro passing through the slips earns. Over a short run of bets ROI is extremely noisy and the number says almost nothing.

Example: One hundred bets of €10 is €1,000 turned over; €50 of net profit is a 5 % ROI.

A bet where the offered price is longer than the true probability of the outcome justifies. It is not a safe bet: a genuine value bet can lose ten times in a row. The edge only shows up across a long series of bets.

Example: A model puts the home win at 50 % (fair price 2.00) while the book offers 2.20. Expected value is 0.50 × 2.20 − 1 = +0.10, or +10 % per unit staked.

The sum of the probabilities that each shot in a match ends up as a goal. The model weighs the shot location, the type of chance, the body part used and the situation it came from. So xG says how many goals a side would score with average finishing, not how many it actually scored.

Example: Five shots worth 0.45, 0.30, 0.12, 0.08 and 0.05 add up to 1.00 xG for the match — even if the side finished it goalless.

The same idea seen from the defence: the summed quality of the chances a side allowed. A low xGA means the back line only lets opponents shoot from difficult positions. The xG minus xGA difference is a simple measure of overall control.

Example: A team averaging 1.4 xG and 0.7 xGA creates chances twice as good as the ones it concedes, even where the league table looks tighter than that.

The points a side should hold based on how its matches actually ran. The xG of both sides converts into win, draw and loss probabilities for each fixture, and those give the expected points haul. A wide gap between real points and xPTS usually signals luck in one direction or the other, and luck evens out.

Example: A fixture priced at 50 % win, 25 % draw and 25 % loss is worth 0.50 × 3 + 0.25 × 1 = 1.75 expected points.

In betting this is usually a synonym for ROI — profit against turnover. Some sources use it for profit against bankroll instead, so it is worth checking which one is meant. Comparisons across sources only work when the definition matches.

Example: Two sources with identical profit can report different yields purely because one divides by turnover and the other by bankroll.