Functional limit theorems for probability forecasts
This paper generalizes the scoring rule for the elicitation of a probability for an event to the score process for scoring probabilities for a collection of events. Precisely, if \(X_ 0, \dots, X_ n,\dots\) is a sequence of \(R^ d\)-valued random vectors, \(\varepsilon\) is a collection of subsets of \(R^ d\), and \(P_{i-1} (E)\) is the elicited probability for the event \(E\) in \(\varepsilon\) given \(X_ 0, \dots, X_{i-1}\), then the quadratic score process is a stochastic process indexed by \(E\), where at each fixed \(E\) in \(\varepsilon\) the process is the following real-valued random variable: \({1 \over n} \sum (\{ X_ i \in E\}-P_{i-1} (E))^ 2\). A functional central limit theorem is obtained for this score process when \(P_{i-1}(E)\) is the conditional probability that \(X_ i\) is in \(E\) given \(\sigma (X_ 0, \dots, X_{i- 1})\). It is shown that the score process converges in distribution to a mixture of Gaussian processes. For the entire collection see [Zbl 0773.00018].
This page was built for publication: Functional limit theorems for probability forecasts
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3139509)