The log-loss score evaluates how well a forecasted probability matches the actual outcome. Instead of simply marking predictions as right or wrong, it measures how confident the forecaster was. Accurate probabilities earn low (better) log-loss values, while incorrect or overly confident predictions receive much higher penalties. This makes the log-loss score a preferred metric in many forecasting and machine-learning competitions.
In prediction markets on platforms like Polymarket, Kalshi, Myriad, and Manifold, log-loss scoring helps analysts understand how well markets anticipated outcomes. When traders adjust probabilities throughout an event, the final log-loss score reflects how closely the market aligned with reality. Markets that incorporate information efficiently and avoid extreme overconfidence tend to achieve stronger log-loss performance.
Log-loss is especially valuable because it aligns with good forecasting behavior: reward calibrated probabilities, discourage unjustified certainty, and highlight where markets responded appropriately to information.
Log-loss scoring provides a precise and fair way to evaluate probabilistic forecasts. It helps analysts determine whether prediction markets produce reliable, well-calibrated prediction markets data.
Log-loss captures both accuracy and confidence. Markets that assign sensible probabilities to events—even when they turn out differently—are scored appropriately, while overconfident mistakes receive steep penalties. This makes log-loss a strong tool for measuring forecast quality using prediction markets data.
Forecasters avoid extreme probabilities unless they’re justified, because log-loss punishes overconfidence. This leads to smoother, better-calibrated probabilities and reduces the chance of dramatic scoring penalties. The result is cleaner prediction markets data and more disciplined crowd forecasting.
Analysts can compare performance across event categories, identify when markets became miscalibrated, and measure how efficiently information was absorbed. High log-loss scores may reveal slow reactions, noisy trading periods, or structural weaknesses. These insights make prediction markets data more interpretable and useful for improving forecasting systems.
A research team evaluates a set of Polymarket political markets by calculating log-loss scores after resolution. They find that markets with stronger liquidity and more consistent participation achieved lower log-loss values, showing better calibration throughout the event cycle.
Evaluating log-loss requires complete, time-stamped probability histories and final outcomes. FinFeed's Prediction Markets API provides structured prediction markets data that developers can use to compute log-loss metrics, compare market calibration, and build tools for evaluating forecasting quality.
