Loading
Four perpetual venues, read from their own indexers rather than from an aggregator. On GMX every open position is counted, 3,795 of them across two chains and not a sample, so the leverage on this page is measured rather than estimated.
The exposure column is never added up. GMX, dYdX and Avantis publish open interest; Ostium’s figure is summed notional over a capped page of open trades and is a floor. Four different measurements under one headline is how the size of this market is usually quoted. It is not quoted that way here.
Forecasting accuracy
A Brier score of 0.25 is what saying “fifty per cent” to everything earns. It is a weak bar here: the forecast scored is the price paid, which is the market’s own estimate, so a score below 0.25 says the markets these agents traded were informative, not that the agents added anything. Whether they did is the price test beside it. On the price test, what they bought won about as often as its price said: no information beyond the market is shown.
By price band, fill by fill: what they bought between 60% and 80% won 84% of the time at an average price of 69%, and the 91 fills bought above 90% at 94% won 68% of the time. The prices are the market’s, so the bands describe the markets they traded as much as the agents; a band can hold many fills on one question.
Buys only: a sell is the opposite statement about a position already held, and scoring it would count one opinion twice with the sign flipped. Forecasts on markets that have not resolved are counted and not scored. The Brier score and the diagram count fills; the price test counts each market outcome once. The forecast is the execution price, which is not necessarily what the trader believed, and outcomes inside one market are not independent of each other. The sample is whatever attributable wallets exist, which is small — a score over a handful of forecasts is a number, not a ranking.
Real-world exposure held through these venues: what the capital is exposed to. Which venues we read at all: what we cover.