◉Hold Desk Open the partner account
paid
Affiliate disclosure. The partner link in the masthead and in the bands beside the copy on this page is a sponsored link to a partner operator, and this site may be paid if you open an account through it, at no extra cost to you. It carries rel="sponsored noopener" and opens in a new tab. That matters on this desk in particular: the subject is the money a gambling business keeps, and this site's own revenue depends on a reader opening an account with one. There is no ranking, no review and no recommendation of any operator anywhere on this site.
Hold Desk / Hold and edge
Figure 03

A measured hold is an average of results, and the edge is a property of a game

The edge on a game is arithmetic: how much of every pound staked the game is built to keep over a long run. The hold on a book of play is an observation: what was actually kept in a period. They converge as the sample grows, and the whole of this page is about what happens before they do.

Figure stub
Sheet
HD-58-04
Subject
Measurement against arithmetic
Edge
a property of a game
Hold
an outcome of a period
gross winWhat was kept: everything staked minus everything returned to players
the holdThe gross win divided by turnover, measured over a period and a whole book of play
net revenueWhat survives the bonus line, and the definition a reader should check first
Direct answerThe house edge is the retention a game is built to take from every pound staked over a very long run. A hold is what was actually retained in a period, across a book of play. They converge as the sample grows: on one day the measured hold can be negative or several times the edge, and over a year it has to sit near it.
The edge is arithmeticThe hold is an observationVariance rules the short sampleA negative hold is possible

Two quantities with one name in common usage

the edge Expected, and knowable in advance

A figure the game's design fixes: the difference between the true odds of a result and the odds the game pays. It is not affected by how many customers played, how much they staked, or which results came in. It is the same on the first spin and the millionth.

the hold Realised, and known only after

Turnover less winnings, divided by turnover. It is what the edge plus variance produced in a period. It is affected by the sample, the stake distribution, the product mix, and by every large win booked against the period's turnover.

Worked example - the same 4% edge, two samples (illustrative) A game with a 4% edge is staked 1,000 times at 1.00: turnover 1,000.00 expected retention: 1,000.00 x 0.04 = 40.00 one large win of 500.00 lands in the sample, and nothing else changes realised gross win: 1,000.00 - 460.00 (ordinary returns) - 500.00 (the win) = 40.00 - 500.00 = -460.00 realised hold: -460.00 / 1,000.00 = -46.0% The same game is staked 1,000,000 times at 1.00: turnover 1,000,000.00 expected retention: 1,000,000.00 x 0.04 = 40,000.00 the same single 500.00 win now costs 0.05% of turnover, not 46% realised hold: approximately 4.0% Nothing about the game changed between the two rows. Only the sample did.

Why a short period can report a negative hold

A hold is a net figure, and a net figure can be negative. When one customer's win is larger than everybody else's losses in the period, the gross win for that period is negative - the operator paid out more than it took in - and a report of that period will show it. This is not an error and it is not evidence of a badly run business; it is what an average looks like before it has averaged.

The practical consequence for a reader is that a hold quoted for a short period says almost nothing about the edge behind it, and a hold quoted for a year says a great deal. It is the same reason a slot machine's published return can be 96% while a session on it loses a player everything: the published figure describes the long run, and a session is not the long run.

How the two converge

After 100 stakeswide
After 10,000 stakesnarrower
After 1,000,000 stakesnear the edge
the band the measured hold can occupy around the same 4.0% edge

The bars shrink as the sample grows, and the edge is the point they shrink towards. Nothing in the mechanism decides how fast they shrink except the spread of the possible results - which is what another desk in the series calls volatility, and what this one needs only as a reason not to read a short hold as a fact about a game.

one

booked, not promised A hold is booked against turnover, not against deposits. A large win by one customer is deducted from the whole period's turnover, regardless of how little that customer staked to win it. This is the single most common way a hold surprises a reader.

two

aggregate, not per game Most reported holds are aggregate across products. A single reported figure can conceal a high-edge game and a low-margin sports book sitting in the same denominator. The mix page unpacks it.

three

not a rate of return on capital A hold is not a return on anything invested. It describes the money staked, not the money a business put at risk, and it says nothing about whether a period was profitable. The lines that decide that are on the pound-of-stakes page.

The honest version of the claim. A game is built with an edge; a business reports a hold. Saying "the casino's edge is 4%" is usually shorthand for a long-run average across a whole product, while saying "the casino held 4% last month" is a statement about one month of results. When the two are used interchangeably, a reader loses the ability to ask the question that matters: on how much play was that measured, and over how long.