LINKS: Prediction Market Turmoil, Breaking the Rules, and Aleatoric Uncertainty

Some links for your Wednesday.

Paper Proposes New Way to Score Election Models… Evaluate Them Like They’re Being Used for Betting!

By: RAJIV SETHI | Source: Markets, Models, and Mamdani

"The method of evaluation is based on a profitability test—a model is said to perform better than the market if a hypothetical trader who believes the model (and is endowed with a budget and some tolerance for risk) makes a profit rather than a loss when updating their portfolio at each model update. This approach can also be used to compare models with each other. Why use a profitability test rather than a conventional measure such as the average daily Brier score? For two reasons. First, imagine a pair of models, one of which perfectly forecasts the other one day ahead. These models produce virtually identical forecasts over a long period of observation, and will have similar accuracy scores under conventional metrics. But one of the two models adds no new information, since its forecasts are themselves perfectly predictable. The profitability test is sensitive to the timing of forecasts (relative to the market) and can yield very different results for the two models in this case."