Machine-Learning-Based Simulation for Estimating Parameters in Portfolio Optimization: Empirical Application to Soybean Variety Selection

Many new seed varieties with traits desirable for different planting environments are developed every year and marketed to farmers. However, farmers lack decision support tools to utilize the vast amount of historical yield performance data to make informed seed variety selection decisions for their individual farms. An informed decision requires accurate estimation of yield performances of seed varieties on the targeted farmland and balancing trade-offs between the expected yield and the risk associated with the seed varieties selected to grow. For that purpose, this paper proposes an analytics framework that integrates machine-learning, simulation, and portfolio optimization to optimally select soybean varieties to grow at the target farm. Using a soybean seed testing dataset collected between 2008 and 2014 by Syngenta, an agribusiness, we choose a machine learning model, which simulates the yield performance of soybean varieties under different plausible weather scenarios at the target farm. The simulated yields are then used to estimate parameters in a portfolio optimization formulation that selects the optimal portfolio of seed varieties to grow at the target farm. Our analysis indicates that an average farmer will gain as much as $177,369 per year in revenue by utilizing the analytics framework. The methodology developed in this research can be applied to variety selection decisions for other crops and influences farming practice positively.