Expected Value, Variance and Risk of Ruin

Expected value tells you whether a decision is worth making. Variance tells you how long you will be unable to tell. Risk of ruin tells you whether you will still be solvent when the answer finally arrives.

Most people who fail at a systematic method do not fail on the first. They get the expected value right and are destroyed by the second two.

Expected value is the average result of a decision repeated indefinitely: every possible outcome, multiplied by its probability, added together.

The critical word is average. Expected value is a property of the decision, not a forecast of the next result. A positive-expectation bet loses constantly. A negative-expectation bet wins constantly. Neither fact says anything about whether the decision was correct, and treating an individual result as a verdict on a decision is the most expensive error available in any of this.

For ordinary casino play the expectation is negative by construction: the game is configured to return less than it takes, so every wager carries a share for the house. That is not a flaw you can outplay. It is the design.

Which means an edge cannot come from playing an ordinary bet better. It can only come from finding a situation in which the expectation has moved — and then betting only in that situation.

Variance measures how widely real results scatter around the average. High variance means long stretches that look nothing like the expectation.

This is the part that ruins people, and it is worth being blunt about why. A thin edge on a high-variance bet can go a very large number of trials before results reliably resemble the maths. During that stretch, a correct method and a broken one produce results you cannot tell apart. There is no observation you can make that distinguishes them, because the distinguishing information does not exist yet.

So the practitioner is left doing something genuinely difficult: continuing to act on a method whose evidence has not arrived, while losing. Almost everyone stops. The ones who do not are usually the ones who understood in advance that this phase was coming and budgeted for it.

Jackpot-style outcomes are the extreme case. A game whose expectation depends heavily on a rare large result has enormous variance, which means the stretch during which nothing appears to work is proportionally longer.

Risk of ruin is the probability of losing your whole bankroll before the edge has time to show. It depends on three things: how large the edge is, how much the results scatter, and what fraction of the bankroll you put at risk each time.

The counterintuitive part is which of the three usually binds. It is almost never the edge. Two people with an identical, genuine edge can have wildly different outcomes purely from bankroll relative to stake, and the one who is underfunded will go broke while being completely correct.

The practical reading: an edge you cannot survive is not an edge. A method that is right and unfunded produces the same result as a method that is wrong.

Given an edge, there is a stake that maximises long-run growth — the Kelly criterion formalises it. Its useful lesson is not the formula but its asymmetry.

Betting less than the optimal fraction grows your bankroll more slowly and is otherwise safe. Betting more than it reduces long-run growth even when the edge is completely real, because drawdowns compound against you faster than gains compound for you. Overbetting a genuine edge is a way to lose money while being right.

In practice most people who apply this deliberately stake a fraction of the optimal amount, accepting slower growth in exchange for a much smaller chance of a catastrophic drawdown — and because the inputs are estimates, and an overestimated edge pushes the optimal stake past the point of safety.

Then you already run this whole framework and the vocabulary is the only thing that differs.

A setup you will not take is a negative-expectation situation you declined. Position sizing is exactly position sizing. A drawdown you sat through is variance. Blowing an account on a strategy that was actually sound is risk of ruin. The discipline of not trading when nothing qualifies is, precisely, the discipline of advantage play.

Two differences are worth knowing. A casino game’s distribution is fixed and published — the rules are certified and do not change because other people have noticed them, which is a luxury markets do not offer. But the edge is capped: you cannot scale a machine-based edge by putting more capital behind it, because the opportunity is limited by how many favourable situations physically exist near you.

High certainty, hard ceiling. That is the honest trade.

Most machines never present a favourable situation at all. They retain nothing between plays, so there is no state that could shift an expectation, and the framework above has nothing to operate on. No amount of sizing or discipline creates an edge that is not there.

A minority of games retain a visible state their own published rules attach consequences to. On those, the expectation of a play depends on that state, and it is readable before you commit anything. Whether a given state is favourable depends entirely on the specific game, version and configuration — which is why this page gives you the framework and not a number. A figure that is right for one cabinet is wrong for the next.

Run the arithmetic · Check what a machine is showing · A fully worked example, free

A qualifying state is not a promise. It describes the condition of a game at a moment in time; the outcome remains uncertain, and a play that qualifies can still lose.

• What advantage play actually is, using card counting as the reference point
• The arithmetic behind a losing session, and what was never your mistake
• Which signals are readable before you sit down, and which are folklore
• What software can and cannot do at a machine, category by category
• Advantage play you can verify yourself, with no machine involved
• What actually happens at the machine, step by step
• Planning the visit: budget, property type and what to arrange first
• What happens after a machine locks up, and the paperwork that follows
• Twelve widely-held beliefs, and which halves of them are true
• How each maker reuses its own design logic across its titles
• Why a bingo-based machine is a different game wearing a slot costume
• Route gaming: how a bar or truck stop floor differs from a casino one
• Which machine to sit at, and what you genuinely cannot know from outside
• The operator's side: house edge, volume, comps and why the offers exist