A0377
Title: A periodic integer-valued time series with an application to fire activity
Authors: Claudia Susana Santos - Polytechnic Institute of Coimbra (Portugal) [presenting]
Isabel Pereira - University of Aveiro (Portugal)
Abstract: Many INteger-valued AutoRegressive models (INAR) are based on the binomial thinning operator to model non-negative integer-valued time series. In real-life events, data with positive and negative integer values can arise. The signed thinning operator allows for negative values both for the series and its autocorrelation function. Focus is placed upon an INAR(1) process with periodic structure and Skellam-distributed innovations. A brief summary of the definition and properties of the signed periodic INAR model of order 1, denoted by S-PINAR(1), is provided. Forecasting is an important topic in time series. Therefore, point forecasts are computed. A simulation study is conducted to give additional insight into the sample behavior of the forecasts. Approximate point forecasts based on the Gaussian approximation are derived and compared. A performance analysis of prediction intervals for forecasts is presented. We conclude with an application concerning fire activity in Portugal.
