{"id":"arxiv/2610.28931","url":"https://arxiv.org/abs/2610.28931","tags":["math.ST"],"score":0,"title":"Boot-AVT: Asymptotically Anytime-Valid Testing with Degenerate U-Statistics via Bootstrap","domain":"math.ST","source":"arxiv","authors":["Jixuan Liu","Long Feng","Liuhua Peng","Zhaojun Wang"],"reviews":0,"subject":"math.ST","summary":"Many nonparametric tests are based on $U$-statistics that are degenerate under the null hypothesis. We develop Boot-AVT, a bootstrap procedure for asymptotically anytime-valid testing with these statistics. Using Brownian scaling and time inversion for the limiting Gaussian chaos process, we show that logarithmically penalized maxima over observed and future sample sizes have a common limiting null distribution. A quadratic multiplier bootstrap estimates the corresponding critical values. We also develop Boot-AVT-Multi for joint testing with multiple kernels. For both procedures, we…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-20","reactions":{}}