Sizing for the jump

An event contract has a fixed end. At some point the event either happens or it does not, and the contract settles at 100¢ or 0¢. Nothing settles in between. Every decision we make about risk starts from that fact.

The mark and the jump

Before resolution we mark positions to market: we value each one at current fair value and record gains and losses as that value moves.¹ Marking is the right way to keep daily books, but it can make a position look calmer than it is. Say we buy YES at 62¢. Over the following weeks the price might wander between 55¢ and 70¢, and the mark will show modest gains and losses along the way. At resolution the contract is worth 100¢ or 0¢. Per contract, the position ends at +38¢ or −62¢, with nothing in between.

That pair of numbers is the terminal exposure, and it is what we size to. For a YES bought at price c, the two outcomes are +(100 − c) and −c. For a NO bought at c, they are the same with the event reversed.

The alternative, sizing to recent price volatility, fails in a specific way. A contract on a monthly data release can trade in a narrow range for days, then move to 100¢ or 0¢ within a minute of the figure coming out. A limit based on the quiet days would allow the largest position just before the largest move the contract will ever make. A limit based on the loss at resolution does not care how calm the price has been.

So every limit we set is a worst-case loss at resolution. Say the limit for one event is $500. Then we can hold 800 YES contracts bought at 62¢, because 800 × 62¢ = $496. Recent price history does not enter that calculation.

One fact, many contracts

Contracts that look separate often hinge on the same fact. A scheduled data release might support a ladder of contracts, one for each threshold the figure could clear. An election might be listed as an outcome set, one contract per candidate, where exactly one will pay and the prices of all of them should add up to about 100¢. A weather reading might settle a contract on whether the daily high clears one level and a second contract on whether it clears another. Each of these is one event carrying several labels.

So we net exposure by underlying event rather than by contract. For each event, we list the outcomes that matter for our positions, work out the combined result of every position in each outcome, and apply the limit to the worst one.

Here is a hypothetical case with round numbers. Two contracts settle on the same city’s daily high temperature. One asks whether the high will be above 85°F, and we hold 1,000 YES on it bought at 50¢. The other asks whether the high will be above 90°F, and we hold 1,000 YES on it bought at 30¢. They trade under different names at different prices, and on their own they look like two modest positions.

Daily high	YES above 85°F, bought at 50¢	YES above 90°F, bought at 30¢	Combined
85°F or below	−$500	−$300	−$800
Above 85°F, up to 90°F	+$500	−$300	+$200
Above 90°F	+$500	+$700	+$1,200

Checked one contract at a time, both positions pass a $500 limit. Checked by event, the worst case is −$800, because both positions lose on the same reading. A limit enforced per contract would have allowed the second trade. A limit enforced per event would have blocked it, since the first position already uses the full $500 in the outcome where both lose.

Netting works in the other direction too. A position that pays in the outcome where the others lose reduces the event’s worst case, and the limit gives it credit. We only give that credit when the contracts settle under the same rules: the same data source, the same time, the same rounding. Two contracts that sound alike can settle on readings from different weather stations, and the difference between them is risk we keep.²

Risk arrives on a schedule

A contract on a monthly release may trade for weeks, but the step from an estimate to a known answer happens in the minute the figure comes out. Every contract on that release reprices at the same moment. An exposure that looks spread across many contracts and many days is really concentrated into a few minutes.

That changes how we prepare. We check each event’s worst case against the moment of release, not against an ordinary trading day. We reduce size and widen quotes ahead of it. And we treat a position as open until the contract actually settles under its rules, however clear the outcome looks from the first headline.

Controls outside the strategy

The strategy code decides what to quote and at what price. A separate layer decides what is allowed. The two run independently, so that a mistake in the strategy cannot switch off the checks meant to catch it.

• Pre-trade checks: every order is checked before it is sent, for its size, for its price against a band around our fair value, and for what the event’s worst case would be if it filled.
• Position caps: limits per contract and per event are enforced by the risk layer, whatever the strategy asks for.
• Kill switches that fail closed: if the risk layer loses contact with the strategy, or cannot confirm that it is behaving as expected, quoting stops.³ The default state is no quotes.
• Stale inputs: each quote depends on a set of inputs (our fair value, the venue’s order book, the reference data the contract settles on). If any of them stops updating within its expected interval, the quotes that depend on it are pulled until it recovers.

That last rule sits alongside what we say in What a two-cent spread pays for about staying in the market through news. Width is how we price risk we can see. A quote built on an input that has stopped updating carries risk we cannot see, and no width covers that.

Footnotes

1. To mark to market is to value a position at its current price or fair value, as if it were closed today. ↩

2. The reverse case matters as much. Two positions that look like a hedge but settle on different sources can both lose, if the two readings land on opposite sides of the threshold. ↩

3. A control fails closed when, unable to tell whether trading is safe, it stops trading rather than letting it continue. ↩