{"id":"arxiv/2610.22742","url":"https://arxiv.org/abs/2610.22742","tags":["stat.ME","math.ST"],"score":0,"title":"Negative-binomial ensemble filtering for discrete-time Hawkes network reconstruction","domain":"stat.ME","source":"arxiv","authors":["Naratip Santitissadeekorn"],"reviews":0,"subject":"stat.ME","summary":"We develop a sequential ensemble filter for reconstructing directed excitation networks from overdispersed multivariate counts observed on a regular time grid. The data-generating model is a discrete-time negative-binomial Hawkes process, for which we prove exact first- and second-moment stability conditions and derive a row-wise sufficient condition. The filter learns the network and the dispersion online, and a row-composition parameter forecast keeps enough ensemble variation to respond to structural change. In synthetic experiments, we demonstrate that a Poisson-based filter provides…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-10-15","reactions":{}}