Spunk Odds Reference

Coin flip guide

Arctic Coin Flip: the arithmetic behind a 1.98x payout

The word Arctic in this address is a leftover label and nothing more. No cold-weather variant of the game exists. What follows is a close look at one number, the 1.98x payout that many coin flip formats quote.

A fair coin would pay exactly 2x

Picture two friends who each put one unit on the table and toss a coin. The winner takes both units, so a winning call returns 2 units in total: the original stake plus one unit of profit. Each side has a 50% chance, and the expected result is zero. Nobody has an advantage, and nobody pays for the service of running the toss.

A service that offers the same toss has to be paid somehow. It does not charge a fee at the door. Instead it trims the payout. In the generic model used on this site, a win returns 2 x (1 - e), where e is the house edge. With a 1% edge that is 1.98x. The coin itself stays fair; only the reward for being right shrinks.

Expected value, line by line

  1. Chance of a correct call: 1/2.
  2. Total return on a correct call: 1.98 units.
  3. Average total return per flip: 0.5 x 1.98 = 0.99 units.
  4. Subtract the unit you staked: -0.01 units, which is -1.00% of the stake.

So the 0.02 that is missing from the payout becomes a loss of 0.01 per unit staked, on average. Half of the flips are wins, and each win is 0.02 short, so 0.02 x 0.5 = 0.01. That is the whole mechanism. There is no hidden bias in the coin and no need for one.

What a 1% trim adds up to

One percent sounds small, and on a single flip it is small compared with the swing of a full unit either way. The table shows the chance of winning several flips in a row, the expected net over that many one-unit flips, and the spread around it.

Coin flips paying 1.98x (edge 1.0%), one unit per flip
FlipsChance all n winExpected net (units)SD of net (units)
150.0000%-0.0100.990
312.5000%-0.0301.715
61.5625%-0.0602.425
120.0244%-0.1203.429
400.0000%-0.4006.261

At 40 flips the expected net is only -0.40 units, while one standard deviation is about 6.26 units. Luck dominates a short session. The edge never goes away, though. It is charged on every unit that is staked, so total turnover, not the number of sessions, decides how much it takes.

Other edges, other payouts

A service could pick another edge. At 2% the payout would be 1.96x and the expected loss 2.00% per unit. At 0.5% it would be 1.990x. The rule is simple: compare the quoted payout with 2.00x, and half of the gap is your average cost per unit.

Is the coin biased to make up the 1%?

Not in this model. The coin is fair and the edge comes only from the reduced payout. A service that also biased the coin would be charging twice.

Could I pick heads every time to avoid the edge?

No. Heads and tails pay the same and win equally often, so the choice of side does not change the expected loss.