{"id":"arxiv/2604.19150","url":"https://arxiv.org/abs/2604.19150","tags":["stat.ME","math.ST"],"score":0,"title":"The General Formulation of Loss-Based Priors for Parameter Spaces","domain":"stat.ME","source":"arxiv","authors":["Cristiano Villa"],"reviews":0,"subject":"stat.ME","summary":"Loss-based priors assign probability mass to parameter values according to the inferential loss incurred when they are excluded from the parameter space, and provide a general solution for discrete parameters. Extending this idea to continuous settings is challenging, as the exclusion of a single point induces no loss. We propose a neighbourhood-exclusion framework in which inferential loss is defined by removing a local region around each parameter value. Under standard regularity conditions, this yields a class of prior distributions driven by the local geometry of the Kullback--Leibler…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-04-21","reactions":{}}