01.Binary outcomes change the arithmetic
In most markets a bad position bleeds and you can cut it. Here, a contract that resolves against you is worth exactly nothing, and resolution can arrive without warning. You can sell before resolution, but only if there is a buyer — and liquidity is thinnest precisely when the news that ruins your position lands. Plan sizing as though the exit may not be there.
- A losing contract settles at zero, not at a reduced price.
- Selling early requires a counterparty, and they disappear on bad news.
- Size on the assumption you cannot get out, then treat an exit as a bonus.
02.Fractional sizing, and why full Kelly is a trap
The Kelly criterion gives a mathematically optimal fraction given your edge and the odds. Its weakness is that it assumes you know your edge, and in prediction markets your edge is an estimate of an estimate. Most practitioners use a fraction of Kelly — a half or a quarter — precisely because being wrong about your edge is more likely than being wrong about the arithmetic.
- Full Kelly is optimal only if your probability estimate is exactly right.
- Half or quarter Kelly trades a little growth for a lot of survivability.
- If you cannot state your edge as a number, you cannot use Kelly at all — size small instead.
03.Correlation is the silent killer
Ten positions across ten markets feels diversified until you notice they all resolve on the same election, or all depend on the same policy decision. Correlated positions are one position wearing several costumes, and the total exposure is what matters. This is the most common way a carefully sized book turns out not to have been sized at all.
- Group positions by what actually determines them, not by market title.
- Size the correlated group as a single position against your bankroll.
- Election and policy markets correlate far more than their titles suggest.
04.Time is a cost you are paying
Capital in a market that resolves in nine months is capital you cannot use for nine months. A position that looks fine in isolation may still be a poor use of the bankroll if it locks funds through a period with better opportunities. Judge long-dated positions on annualised return against alternatives, not on absolute payoff.
- Long-dated positions carry opportunity cost that the payoff must justify.
- Compare on annualised terms, not on the headline return.
- A 20% return over a year is a very different trade from 20% over a week.