Original title: Moments for self-normalized partial sums
Muneya Matsui, Thomas Mikosch, Olivier Wintenberger · September 2026
When numbers in a sequence fluctuate wildly with rare, extreme spikes, calculating typical statistics becomes difficult. The total sum can behave unpredictably unless you balance it out. Researchers analyzed stationary sequences where extreme values follow a heavy-tailed pattern with tail indexA parameter that determines the heaviness of the distribution tails; here less than 2 indicates heavy tails. ___ < 2. They scaled the running sums by other measures from the sequence, such as studentized values or the highest number recorded. This self-normalization balances the extremes by dividing the total by its own biggest parts. The researchers checked if these balanced ratios have well-behaved averages, called finite moments. They found that for some sequences, the limiting ratios possess all moments. However, other sequences yield infinite moments instead. This difference happens because extreme values can group together into clusters.
Imagine you are tracking the daily catch of a fisherman over a month. Most days, the catch is moderate, but occasionally, there are days with huge hauls. If you divide the total catch for the month by the best day, this ratio might seem straightforward, but if the huge hauls happen in groups, the ratio could become so erratic that its average doesn't stabilize. In this paper, the sequences are like the daily catch, and extremal clusters are these grouped huge hauls. The ratios studied are similar to that total divided by the maximum, and the paper shows that clustering can cause certain moments of these ratios to be infinite.
For researchers working with heavy-tailed data, knowing if moments of ratios exist is crucial for reliable statistical inference. This affects fields like risk analysis where extreme values are common and models must handle them correctly.
The study is limited to regularly varying stationary sequences with tail index less than 2. It does not address other types of distributions or non-stationary processes. Also, it only examines the existence of moments, not other statistical properties or applications.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Moments for self-normalized partial sums — Muneya Matsui, Thomas Mikosch, Olivier Wintenberger
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.