Reading a Prediction Market

Before any strategy, you need to read the screen correctly. A prediction market board shows the same objects as an equity book, but the units are probability, and that changes what counts as expensive.

The board

A typical market shows Yes bid at 61, Yes ask at 63, last trade 62, with some visible depth at each level. Three immediate readings:

• The mid, $\frac{61 + 63}{2} = 62$, is the board's central probability estimate, the number to quote when someone asks "where is this market".
• The spread, $63 - 61 = 2$ cents, is two full probability points. Crossing it, rather than joining the far side and waiting, costs the full two points of expected value.
• Depth tells you how much conviction stands behind those prices. Fifty contracts at the bid is an opinion; fifty thousand is a wall.

The last trade is the least informative number on the board. In a thin market it can be hours old and a point or more from the current mid; the standing quotes are the live information.

Spreads are probability points

A 2 cent spread means something entirely different at different price levels. At a 50 cent mid it is a $4\%$ round-trip cost on a coin flip, wide but survivable. At a 95 cent price, the Yes side's total possible profit is 5 cents, so a 2 cent spread consumes $40\%$ of the maximum payoff. Markets near the extremes are structurally expensive to trade even when they look tight, one of several reasons the tails behave badly, which the longshot lesson takes up properly.

Thin markets and stale prices

Most prediction markets are small. A market with a handful of resting contracts can be moved several points by one modest order, and its "price" is barely a price at all. Checks worth making a habit before reading any board as a forecast:

1. Is there two-sided depth, or one lonely bid against an empty ask column?
2. When did the last trade actually print?
3. Do Yes and No quotes add up sensibly, or has one side gone stale?
4. How close is resolution? Time compresses everything: the same 2 point spread that was noise a month out is an enormous disagreement an hour before settlement.