Elo Rating
A skill-rating system borrowed from chess. In LDBD it starts at 1500 and moves a predictor's score up or down by how far the result beat the asset's prior probability.
Formula
expected = base_rate_up (if you predicted up) expected = 1 − base_rate_up (if you predicted down) skill_delta = K × (actual − expected) # actual = 1 if correct, else 0 K by timeframe: 1d=32, 1w=24, 1m=16, 6m=12, 1y=8
Elo ratings were invented for chess, as a way to capture the skill gap between two players in a single number. The physicist Arpad Elo came up with it, and the core idea is simple: before a game you use each player's rating to compute a win probability, then you nudge the ratings depending on whether the result beat that expectation. Beat a strong opponent and your rating jumps; beat someone you were always favored to beat and it barely moves.
LDBD's Skill Rating borrows that machinery for forecasting. Everyone starts at 1500, and the role of the 'opponent' is played by base_rate_up, the prior probability that this asset rises over this timeframe. If you called it up, your expected score is base_rate_up; if you called it down, it's one minus that. Getting a habitually-rising asset right earns almost nothing because you were expected to, while nailing the unfashionable side pays much more.
Skill Rating is a secondary metric, not the headline rate on the leaderboard. The interface shows it as a per-bet average rather than a running total, and the K-factor is larger on shorter horizons, so a single call there swings the score more. The simulator below lets you dial the asset's prior, your predicted direction, and the timeframe to watch the points move.
Example
If an asset's base_rate_up is 0.7, calling the popular 'up' side and being right earns only 32 × (1 − 0.7) = +9.6 on a 1-day bet. Calling 'down' and being right earns 32 × (1 − 0.3) = +22.4 — more than double, for going against the obvious.
How LDBD uses it
On LDBD, Elo shows up as the secondary Skill Rating. Where the headline rate measures the size of your gains, Skill Rating measures whether you beat the prior: expected comes from base_rate_up, so an obvious call earns little. When an asset has fewer than 100 historical price samples, base_rate_up falls back to the sector average. It updates each prediction via skill_delta = K × (actual − expected), and the UI displays a per-bet average, (skill_rating − 1500) / (resolved_count + 100).
FAQ
How is Skill Rating different from the leaderboard's main rate?
Rate annualizes your directional log return to measure how much you made; Skill Rating measures how much you beat the prior. A big winning call still earns little Elo if the direction was the obvious one.
Why is K larger for shorter timeframes?
Short-horizon calls resolve quickly and pile up, so a bigger K (32 for 1-day) lets the rating respond fast. Year-long calls are rare and slow, so a smaller K (8) keeps one lucky outcome from swinging the score.