What a Positive or Negative Expected Value Means

You sit down with $100 and a bet that has a positive expected value of $0.50 per play. After 18 consecutive losses, you have $82 left. The math says you are still ahead over the long run, but your bankroll does not care about the long run right now. Expected value is the long-run average outcome of a random event, calculated as the sum of each possible outcome multiplied by its probability. It is not the outcome you expect on a single trial. A positive expected value tells you the average direction over many repetitions, not what happens next.

The sign of the expected value is the first filter. A positive expected value means you average a gain per trial over many plays. A negative expected value means you average a loss. A zero expected value means the game is fair: you break even on average. Most casino games have a negative expected value. American roulette has a house edge of 5.26%, so every $1 bet has an expected value of −$0.0526. A fair coin flip with even money has zero expected value. The table below shows the three cases side by side with real numbers.

Sign	E[X] per $1 Bet	Long-Run Outcome	Example
Positive	$0.10	Average profit of $0.10 per play over thousands of trials	Blackjack with perfect basic strategy and favourable rules
Zero	$0.00	Break even on average; net gain near zero after many trials	Fair coin flip, win $1 on heads, lose $1 on tails
Negative	−$0.0526	Average loss of $0.0526 per play; bankroll declines over time	American roulette (double-zero), any single-number bet

The single most common mistake in expected value interpretation is treating a positive expected value as a guarantee of profit on the next play. It is not. The Law of Large Numbers says the sample average converges to the expected value as the number of trials increases. That convergence is slow. A positive-EV bet with a small edge, say $0.01 per $1 bet, needs tens of thousands of trials before the average outcome reliably exceeds zero. One losing streak can wipe out a small bankroll before the math catches up.

OpenIntro Statistics (4th ed.) section 3.4 defines random variables as numerical outcomes of random processes and gives the discrete expected value formula E[X] = Σ x·P(X=x). That formula produces a single number. It does not tell you what happens on trial number 7. It tells you what happens on average over many trials.

Two bets can have the same positive expected value and radically different risk. Consider Bet C: a 50% chance to win $2 and a 50% chance to lose $1. Expected value = (0.5 × $2) + (0.5 × −$1) = $1.00 − $0.50 = $0.50. Bet D: a 1% chance to win $51 and a 99% chance to lose $0. Expected value = (0.01 × $51) + (0.99 × $0) = $0.51. Both have about $0.50 positive expected value. Bet C has a standard deviation of $1.50; Bet D has a standard deviation of about $5.07. The variance formula from Blitzstein & Hwang (2nd ed.) ch. 5, Var(X) = E[(X − μ)²] = E[X²] − (E[X])², confirms that Bet D is more than ten times as variable. The same EV with different variance means a very different chance of ruin.

A simulation of 1,000 trials for Bet C (EV $0.50, SD $1.50) and Bet D (EV $0.50, SD $5.07) shows the divergence. After 1,000 plays of Bet C, the cumulative profit clusters around $500 with a narrow spread. After 1,000 plays of Bet D, the cumulative profit ranges from −$200 to $1,200. Both converge to the same average per trial, but the path is far bumpier for the high-variance bet. The practical takeaway: a positive expected value is necessary but not sufficient. You need to know the standard deviation to judge whether you can survive the swings.

Compute the Kelly Fraction Before You Bet

A positive-EV bet with high variance can bankrupt a small bankroll before the Law of Large Numbers kicks in. This is not a theoretical edge case. Kelly (1956) "A New Interpretation of Information Rate" derived the optimal fraction of bankroll to bet to maximize long-run growth: f* = (p(b+1) − 1) / b, where p is the win probability and b is the net odds received. Betting more than the Kelly fraction increases the risk of ruin. Betting less reduces growth but protects the bankroll.

For Bet C above (50% chance to win $2, 50% chance to lose $1), the Kelly fraction is (0.5 × (2+1) − 1) / 2 = (1.5 − 1) / 2 = 0.25. A player with a $100 bankroll should bet at most $25 per trial. Betting the full $100 leads to a 50% chance of losing everything on the first play. Positive EV does not protect against a single loss.

Understand Why You Accept a Bad Bet

The expected value model assumes linear utility: each dollar has the same value whether you have $10 or $10,000. Real humans are risk-averse. A $1,000 loss hurts more than a $1,000 gain helps. Blitzstein & Hwang (2nd ed.) on utility defines a concave utility function for risk-averse individuals. Insurance premiums have a negative expected value; the insurer's house edge is built into the price. A homeowner pays $1,200 a year for fire insurance on a 1-in-200 chance of a $200,000 loss. The expected value is (0.005 × −$200,000) + (0.995 × $0) = −$1,000, minus the premium of $1,200, for a net expected value of −$2,200. The homeowner accepts this because the utility of avoiding a catastrophic loss outweighs the certain premium cost.

This is why the old advice "if the game is fair, take it" is wrong for anyone with a finite bankroll who is not risk-neutral. A fair game with zero expected value is still risky. The correct decision depends on variance, bankroll size, and utility. A millionaire can accept a fair coin flip for $1,000; someone with $2,000 in savings should not.

A fair game has an expected value of exactly zero. Over many trials, the cumulative profit hovers near zero. The Law of Large Numbers ensures the average approaches zero, but the total profit can wander far from zero before converging. A fair coin flip for $1 has zero EV. After 1,000 flips, the probability of being ahead by $20 or more is about 31%. After 10,000 flips, it drops to about 5%. The long-run convergence is real but slow.

Casino games are never truly fair. The house edge ensures a negative expected value for the player. Even games that appear fair, a sports bet with no vig, a lottery with a massive jackpot, have a built-in negative EV once the bookmaker's commission or the tax and split risk is accounted for. The only exceptions are promotional offers, counting in blackjack, and rare mispriced bets.

Bet Only What the Kelly Fraction Allows

Positive-EV investing is called "the core of value investing," and the principle "if the game is fair, take it" applies under positive EV. Both are wrong. Positive expected value investing is not value investing; value investing focuses on buying assets below intrinsic worth, which is a separate concept. And a fair game (zero EV) should not automatically be taken, because variance and utility matter. A positive expected value does not mean you should bet your entire bankroll, and a zero expected value does not mean the bet is harmless.

The correct action on a positive expected value result: compute the Kelly fraction, compare it to your bankroll, assess your risk tolerance, and bet no more than the Kelly fraction. If you cannot afford the variance, skip the bet even if the EV is positive. If the bet is negative EV, avoid it unless you value the entertainment or the utility of the loss is trivial relative to your bankroll.

Expected value suits anyone making repeated decisions under uncertainty: a gambler comparing casino games, an investor evaluating a sequence of bets, a project manager using EMV in a decision tree, or a statistics student computing EV from a probability table. It does not suit anyone who needs a guarantee on a single trial. If you want to know whether you will win this specific hand or whether this specific lottery ticket is a winner, expected value cannot answer that. The long-run average is not a prediction for one event.

The single thing that most often goes wrong: a bettor finds a positive-EV opportunity, bets too large a fraction of bankroll, and goes bust before the long run arrives. The Kelly fraction exists precisely to prevent that failure.