Search papers, labs, and topics across Lattice.
This paper addresses the problem of forecast instability in multi-step-ahead time-series forecasting, where forecasts for the same target period change as new observations become available. They propose a method to jointly optimize forecast quality and stability by using regression splines parameterized by a neural network to generate stabilized forecasted conditional quantile functions. Empirical evaluations on two datasets demonstrate that the method effectively reduces forecast instability without significantly sacrificing forecast quality, and allows for targeted stabilization of specific parts of the forecast distributions.
Stop your time-series forecasts from flip-flopping: this method lets you directly control the trade-off between forecast accuracy and stability, even focusing stabilization on critical parts of the distribution.
Multi-step-ahead forecasts are often updated as new observations become available, since shorter forecast horizons typically improve forecast quality. However, such improvements come at the cost of forecast instability, i.e., variability in forecasts for the same target period. This instability can trigger costly changes to plans formulated based on the forecasts and may erode trust in the forecasting system. In this work, we integrate forecast stability alongside forecast quality into the training of distribution-free probabilistic time-series forecasting models, allowing us to control this trade-off. We propose a method for generating stabilized forecasted conditional quantile functions using regression splines parameterized by a neural network. This approach enables joint optimization of quality and stability, as it allows us to directly penalize dissimilarities arising from forecast updates. Furthermore, it allows assigning varying importance to stabilizing different parts of the forecast distributions (e.g., central parts vs. tails) to focus on the parts most relevant for the intended downstream use (e.g., the upper tail for inventory management). We empirically evaluate the proposed method on two datasets with different statistical properties and show that it can effectively reduce forecast instability without a substantial loss in forecast quality, and that it can target stabilization effort toward specific parts of the forecast distributions.
