LATE AND EARLY INDICA RICE’S PRICE FORECASTS THROUGH NEURAL NETWORKS

Building price forecasts of agricultural commodities plays a significant role in decision making processes of market participants and policy design/implementations of policymakers. Constructing price forecasts of rice is of particular importance as it serves as a strategic food resource across the globe. In this work, we address such a price forecast problem based on weekly price indices of late and early indica rice in the Chinese wholesale market during the period spanning April 1, 2011–September 13, 2019. We utilize nonlinear auto-regressive neural network models to facilitate forecasting and examine more than a hundred model settings in the fields of the adopted model training algorithm, the number of hidden neurons used, the number of delays utilized, and the ratio employed for data segmentation into different phases. As a result, we construct relatively simple models for price forecasts of the two commodities with stable and accurate forecast performance. Particularly, for the training, validation, and testing phases, the model built leads to relative root mean square errors of 1.08% (0.91%), 1.09% (0.89%), and 1.07% (0.89%), respectively, for the price of late (early) indica rice. The models here could be used to help decision making and policy design/implementations.