Methodology for automated identifying food export potential

Abstract

The food and agriculture could be a driver of the economy in Russia if intensive growth factors were mainly used. In particular, it is necessary to adjust the food export structure to fit reality better. This problem implies long-term forecasting of the commodity combinations and export directions which could provide a persistent export gain in the future. Unfortunately, the existing solutions for food market forecasting tackle mainly with short-term prediction, whereas structural changes in a whole branch of an economy can last during years. Long-term food market forecasting is a tricky one because food markets are quite unstable and export values depend on a variety of different features. The paper provides a methodology which includes a multi-step data-driven framework for export gain forecasting and an approach for automated evaluation of related technologies. The data-driven framework itself uses multimodal data from various databases to detect these commodities and export directions. We propose a quantile nonlinear autoregressive exogenous model together with pre-filtering to tackle with such long-term prediction tasks. The framework also considers textual information from mass-media to assess political risks related to prospective export directions. The experiments show that the proposed framework provides more accurate predictions then widely used ARIMA model. The expert validation of the obtained result confirms that the methodology could be useful for export diversification.