How the markets work

Each paper has a market on its decision, with two outcomes: Accept, if the paper is accepted in any form (oral, spotlight or poster), and Reject. A withdrawn or desk-rejected paper counts as Reject. Each market’s contract, on its page, says exactly how it resolves.

You buy shares in an outcome. When the decision is published, every share of the outcome that happened pays 1 $rep; every other share pays nothing.

So an outcome’s price is the market’s probability for it: a share at 30% costs about 0.30 $rep and pays 1 $rep if it wins. The two prices always add up to 100%, and the “accept” figure shown for a paper is the price of Accept.

Trading is in reputation, $rep, not money. Everyone starts with 1,000 $rep once they have confirmed an institutional email address. There is no way to buy more: the only way to gain is to be right before everyone else is.

You say how much to stake; the trade box shows how many shares that buys, as what it pays if the outcome wins. You always trade with the market maker (§5), so there is never anyone to wait for.

Buying moves the price (2). Each share you buy makes the next one dearer, so a large stake pays a higher average price than the one shown before you bought — the box shows the price before and after your trade.

The cost you are shown is a limit. If someone else trades first and the price moves against you, your order is refused rather than filled at a worse price. If it moves in your favour, you pay less.

You can sell shares you hold, in part or in full, at any time until trading closes. You cannot sell shares you do not hold: there is no shorting. To bet against acceptance, buy Reject.

Selling moves the price down as you sell, so selling a holding pays less than its shares times the current price. That is why your positions show a “current value” — what selling everything now would actually pay — and not shares × price. Right after a buy it is a little below what you paid: buying and immediately selling back loses a little, never gains.

Prices are set by an automated market maker, Hanson’s logarithmic market scoring rule (LMSR) [1, 2]. It always quotes a price for any outcome, and the price depends only on how many shares of each outcome have been bought so far. Write $q_i$ for the shares of outcome $i$ bought so far, out of $n$ outcomes. The market maker keeps the cost function

and the price of outcome $i$ is how fast that cost grows with its shares,

which is always between 0 and 1, and sums to 1 over the outcomes. Buying $\Delta$ shares of $i$ costs the difference in (1) before and after; selling is the same with $\Delta$ negative, and pays back that difference. Costs are rounded to a millionth of a $rep, always in the house’s favour.

The depth $b$ — how far a given stake moves the price — is sized when the market is created, from the number of traders expected and the starting balance, so no single trader can pin a price on their own. It is then fixed for the life of the market. The house funds each market maker with

the most it can lose however the market resolves [1], so every winning share is always paid.

Trading closes at the date shown on the market. When the decision is published, the market is settled on the outcome that happened, with a link to the evidence where there is one: each of its shares pays 1 $rep into your cash, and every other outcome’s shares are worth nothing.

Your net worth is your cash plus what selling every holding right now would pay (§4). It is never shares × price, which would let a trader show a profit just by pushing up the price of what they hold.

The leaderboard ranks everyone on that net worth, or on profit from settled markets alone.