Optimal Alarm Systems for Count Processes

Abstract

In many phenomena described by stochastic processes the implementation of an alarm system becomes fundamental to predict the occurrence of future events. In this work we develop an alarm system to predict whether a count process will upcross a certain level and give an alarm whenever the upcrossing level is predicted. We consider count models with parameters being functions of covariates of interest and varying on time. The paper presents classical and Bayesian methodology for producing optimal alarm systems. Both methodologies are illustrated and their performance compared through a simulation study. The work finishes with an empirical application to a set of data concerning the number of sunspot on the surface of the sun.

Alarm system for insurance companies: A strategy for capital allocation

“…Different approaches have been proposed so far in the literature, defining random or deterministic alarm times; see e.g. [15], [18] and references therein, [8], [5]).…”

Integer-valued autoregressive processes with periodic structure

“…Motivation to include discrete data models comes from the need to account for the discrete nature of certain data sets, often counts of events, objects or individuals. Examples of applications can be found in the analysis of the number of rainy days (Cui and Lund, 2009), time series of count data that are generated from stock transactions (Quoreshi, 2006) where each transaction refers to a trade between a buyer and a seller in a volume of stocks for a given price, statistical control process (Weiß, 2009), telecommunications (Weiß, 2008), and also in the analysis of optimal alarm systems (Monteiro et al, 2008), experimental biology (Zhou and Basawa, 2005), social science (McCabe and Martin, 2005), and queueing systems (Ahn et al 2000).…”

Thinning-based models in the analysis of integer-valued time series: a review

“…For this purpose, Brännäs (1995) and Monteiro et al (2008) considered the INAR(1) model in (2.1) and introduced the effect of explanatory variables through the thinning parameter˛. Usually, the conventional specifi-cation˛t = 1/(1 + e y t ω ) with y t = [y t,1 · · · y t,l ] and ω = [ω 1 · · · ω l ] , is adopted.…”