Precision score

​

How it’s calibrated

​

Formulas (for verification)

• Log-return error: $\varepsilon_i = \bigl| \ln(x_i) - \ln(y) \bigr|$, where $x_i$ is your estimate and $y$ is the outcome.
• Effective volatility $\sigma_{\text{eff}}$: an average of log returns over a rolling 30-minute window, weighted toward recent data (EWMA). Each new return updates it as $\sigma_{\text{eff},t} = \alpha \times |r_t| + (1 - \alpha) \times \sigma_{\text{eff},t-1}$, where $r_t$ is the latest log return and $\alpha$ sets how quickly older data fades out.
• Sensitivity: $\lambda = \ln(2)/\sigma_{\text{eff}}$.
• Precision score: $\mathrm{PS}_i = \max\!\left(100,\; 1000 \times \exp\!\left(-\lambda \times \varepsilon_i\right)\right)$.

​

Leaderboards