Search papers, labs, and topics across Lattice.
This paper introduces ProbRes, a novel post-hoc probabilistic calibration method designed for time-series forecasting that effectively incorporates volatility dynamics to manage heteroskedastic data. By employing two architecture-agnostic modules to model conditional mean and volatility separately during training, ProbRes generates predictive distributions through resampling normalized residuals at inference. Experimental results on synthetic and real-world datasets validate that ProbRes not only captures predictive distributions accurately but also produces well-calibrated prediction intervals, outperforming existing methods.
ProbRes achieves robust probabilistic forecasting by explicitly modeling volatility dynamics, leading to superior calibration of prediction intervals in financial applications.
Probabilistic time series forecasting has attracted increasing attention in financial applications due to the need to quantify risk and uncertainty in future observations. We propose ProbRes, a post-hoc probabilistic calibration method that explicitly learns and incorporates volatility dynamics into probabilistic forecasting, enabling effective handling of heteroskedastic data. During training, ProbRes employs two architecture-agnostic modules to separately model the conditional mean and conditional volatility. At the inference stage, it generates predictive distributions by resampling normalized residuals. ProbRes is applicable to both univariate and multivariate time series and remains robust under a wide range of error distributions, including non-Gaussian innovations with conditional heteroskedasticity. Theoretical results demonstrate ProbRes's validity and experiments on both synthetic and real-world datasets show that ProbRes accurately captures predictive distributions and produces well-calibrated prediction intervals.
