Forecasting the number of vehicles thefts in Campinas/Brazil using a Generalized Linear Autoregressive Moving Average model

By definition, thefts are considered the act of taking away other people's mobile possessions for personal use or for others, affecting crime rates, economic indicators and enabling recent studies to create risk zones in society, contributing to insurance pricing in actuarial methods. This paper analyzes the number of vehicle thefts of 38 locations near Campinas/São Paulo, Brazil, using a GLARMA(p,q) model with Poisson and Negative Binomial response. The main feature of GLARMA(p,q) is to consider the peculiarities of counting data as high dispersion. As a result, it was possible to verify the adequacy and usefulness of the model for counting data. With specific techniques for estimating time series related to the public security area, patterns can be better understood, revealing relevant information that can be added to decision-making processes to direct public policies.