Multivariate analysis and neural networks application to price forecasting in the Brazilian agricultural market

Aplicação de análise multivariada e redes neurais para previsão de preços no mercado agrícola brasileiro

ABSTRACT:

The purpose of this study is to apply the methodology proposed by PINHEIRO & SENNA (2015) to a set of agricultural products traded in Brazil. The multivariate and nonlinear character of this methodology has shown to be suitable, as compared to the neural network model, since it allows for a better predictive performance. Results obtained in an out-of-sample period, by using the calculated error and statistical test, confirmed this statement. This study will be useful to farmers as price forecasting based on their tendency is relevant.

Key words:
neural networks; multivariate analysis; agricultural products; forecasting