Members of the Delphi Group have developed an online method shown to improve calibration of forecasts.
MultiQT, the multiple-level quantile tracker, is a generic method for recalibrating quantile forecasts. It is lightweight and easy to implement (requiring ~10 lines of code): using a forecaster’s track record, it adjusts the forecasts prospectively to improve coverage at every pre-specified quantile level. Critically, it does so while ensuring that the quantile forecasts remain ordered, e.g., the 0.5-level quantile forecast will never be larger than the 0.6-level forecast.
Experiments on 15 arbitrarily selected death forecasters from the Covid-19 Forecast Hub show that MultiQT significantly improves calibration of each one. Meanwhile, predictive accuracy according to WIS is mostly unchanged (MultiQT sometimes makes this a bit better, or a bit worse).
Read more about MultiQT in their preprint: https://doi.org/10.48550/arXiv.2512.23671