Gopinath (Gopi) Munisamy is Professor of Agricultural and Applied Economics at the University of Georgia (UGA), Athens, USA, where he conducts research and teaching in International Trade and Agricultural Policy. In his interview to IAMO, he outlines the role of machine learning and artificial intelligence in trade forecasting.
IAMO: How are machine learning methods used in trade forecasting today? What is their forecast accuracy?
Gopinath Munisamy: Machine learning (ML) and artificial intelligence (AI) have been around for a long time in financial markets, e.g. forecasting stock prices and executing stock trades. The economics discipline is just beginning to realize the potential of ML methods to decipher complex relationships in any application – environment, agriculture, labor, energy, trade and others. The few people that work on trade applications of ML have mostly used methods in the descriptive and predictive domain, but we are making progress toward using techniques from the prescriptive domain as well. Common methods for description and prediction include decision trees for both quantitative and categorical data along with bagging, boosting, random forest and extra-tree regression, and neural networks.
In the few applications to agricultural trade, we find the ML forecast accuracy to be above 65% compared to about 35% in some traditional approaches e.g. a combination of model and expert judgements. More importantly, with ML methods we are able to explain factors underlying forecasts.
IAMO: What are the main challenges of applying machine learning techniques in trade?
Gopinath Munisamy: Organizing the large volume of data for applying ML algorithms is a major challenge. ML methods are becoming standard black boxes in advance programming languages (Python). So, it is absolutely critical to clearly identify the research objective, choose the appropriate ML domain and organize the data prior to any application. With ML/AI, especially deep learning, researchers need state-of-the-art computing equipment with high speed and memory, and lots of storage.
IAMO: In your opinion, how will ongoing digital transformation affect the structure of international and regional trade globally? What do you think are the biggest benefits and risks of digital transformation of the food value chain globally?
Gopinath Munisamy: Within countries, ML and AI are providing timely and critical information to manage supply and demand, especially in North America, Europe and Far-East. Better serving customer demand and improved sourcing of supplies are lowering transaction costs including storage. Improved efficiency via this digital transformation is an important source of growth. Across countries, the challenge is that we do not create a two-tier system, i.e. those that can afford ML and AI versus those that don’t even have the human capital to understand and implement these strategies in the marketing/trade dimension.
ML and AI are leading innovations at the producer and processor levels as well. New varieties, chemicals and management techniques (e.g. AI based precision agriculture) are also creating economic opportunities. However, these technologies come with high fixed costs and, again, there is the real danger of leaving behind the majority of farmers and processors who cannot afford these technologies.