Original title: Scalar on Function Absolute Relative Error Regression: Functional Time Series Case
Fatimah A. Almulhim, Mohammed B. Alamari, Ali Laksaci · September 2026
Predicting numbers over time is difficult when outcomes swing wildly or contain extreme errors. Standard methods often struggle because they treat mistakes the same regardless of size. The researchers solved this by introducing a new estimator for predicting a single number from complex curves over time. It works by minimizing Least Absolute Relative Error, or LARE. Instead of measuring pure error distance, this method sizes each error relative to the true value being predicted. Because it focuses on these relative deviations, the approach reduces the impact of extreme points and handles changing variance much better than older methods. The researchers proved its mathematical behavior over time, which allows people to build confidence intervals and test hypotheses. Simulations and real data confirmed it is simple, stable, and accurate.
Imagine judging how good your friend's guesses are at how much things weigh. If you only look at the raw error, a miss that is small compared to the true weight looks the same as a miss that is large compared to it, so sloppy guesses on light objects get treated the same as good guesses on heavy ones. So you switch to judging by proportion: a miss that is a large fraction of the true weight is bad, while a miss that is a tiny fraction of it is fine. In this paper, the same idea is applied to predicting a number from a curve: instead of minimizing raw differences, the method minimizes the error as a fraction of the true answer, so light and heavy cases are judged fairly.
Anyone forecasting quantities that swing across very different scales can get predictions that are not dominated by the biggest values. Because the method treats a miss as a fraction of the true answer, it stays accurate when data has outliers or changing spread. That makes forecasts more reliable in messy real-world settings.
The abstract doesn't state limitations directly. It focuses on introducing the estimator and deriving its asymptotic distribution, then supports it with simulations and one real dataset. It does not discuss how sensitive the method is to the choice of semi-metric, how it behaves when the functional time series assumption is violated, or how it scales to very large datasets.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Scalar on Function Absolute Relative Error Regression: Functional Time Series Case — Fatimah A. Almulhim, Mohammed B. Alamari, Ali Laksaci — Symmetry — doi:10.3390/sym18091563
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.