As the price of soybean affects the soybean market development and food security in China, its forecasting is essential. A quantile regression-radial basis function (QR-RBF) neural network model is introduced in this paper. The model has two characteristics: (1) using quantile regression models to describe the distribution of the soybean price range; and (2) using RBF neural networks to approximate the nonlinear component of the soybean price. In order to optimize the QR-RBF neural network model parameters, a hybrid algorithm known as GDGA, based on a combination of the genetic algorithm (performing a global search) and a gradient descent method (performing a local search), is proposed in this paper. Data regarding the monthly domestic soybean price in China were analyzed and the results indicate that the proposed hybrid GDGA is effective. Furthermore, the results suggest that the influencing factors of soybean price vary at different price levels. Money supply and port distribution price of imported soybean were found to be important across a range of quantiles; output of domestic soybean and consumer confidence index were important only for low quantiles; and import volume of soybean and consumer price index were important only for high quantiles.
Bibliographical notePublisher Copyright:
© 2018 Elsevier B.V.
- Genetic algorithm
- Gradient descent
- Quantile regression-radial basis function (QR-RBF) neural network
ASJC Scopus subject areas
- Agronomy and Crop Science
- Computer Science Applications