A combined forecast model: an application
to the monthly series on dengue reports
in Pernambuco state - Brazil

Dirac Moutinho CORDEIRO[1]

Gauss Moutinho CORDEIRO[2]

§       ABSTRACT: A combined forecast model is developed to explain the behaviour of monthly dengue reports in Pernambuco state, Brazil. Bunn's proposal (1985) is used. Two punctual forecasts, the best in terms of efficient dimension - RMSE (Mean Square Erros), were linearly associated to achieve the best forecasts by statistical means, residual analysis and perceptual reduction of unknown combined variables. It is observed that all the methodological approaches to the series show a dramatic reduction in report forecasts as compared with those of 1998 - the year when a generalised epidemics occurred. To our surprise, 1998 had been the year of the largest water rationing in the previous thirty years. The results of forecasts are stable and, in this case, it is possible toextrapolate their behaviour to a stationary process with few oscillations. The series tends to a controlled epidemics.

§       KEYWORDS: Bayesian approach; beta distribution; Box and Jenkins; Bunn;combined forecast model; dengue; linear dynamic model; Monte Carlo error; Winters.

[1]Departamento de Matemática, Escola Politécnica, Universidade de Pernambuco - UPE, CEP 50171-900, Recife, PE, Brasil. E-mail: dirac@emtu.pe.gov.br

[2] Departamento de Física e Matemática, Universidade Federal Rural de Pernambuco - UFRPE, CEP 50171-900, Recife, PE, Brasil. E-mail: gausscordeiro@uol.com.br