Resumo

SANTANA CONTRERAS, Juan Camilo; CAMARO, Álvaro Andrés; CASAS HENAO, Arnoldo  e  JIMENEZ MENDEZ, Édgar. An empirical study of neuronal networks' predictive ability in forecasting colombian inflation: an alternative methodology. Innovar [online]. 2006, vol.16, n.28, pp.187-198. ISSN 0121-5051.

Evaluating the prediction ability of neuronal networks (Box-Jenkins' SARIMA, exponential smoothing and varying coefficient regression models) is interesting in forecasting Colombian inflation. Such knowledge is fundamental in designing economic policy and strategic investment programmes in both the public and private sectors. An application forecasting future values from a series of Colombian inflation shows that neuronal networks supported, by non-observable components, could give more precise forecasting compared to traditional Box-Jenkins', exponential smoothing and flexible square minimum methodologies. The results also revealed that forecasting combinations making use of neuronal networks tended to provide better predictions.

Palavras-chave : multi-layer perception; SARIMA models; exponential smoothing; flexible square minimums; forecasting combination; non-observable components.

        · resumo em Português | Espanhol | Francês     · texto em Espanhol     · Espanhol ( pdf )