|
AN AMBIGUITY FRAMEWORK TO INFORM OPTIMAL DATA ALLOCATION FOR DATA-DRIVEN OPTIMIZATION:
ANALYSIS AND APPLICATION TO AGRIBUSINESS
BY: Saurabh Bansa and Ying He
Dow Chemicals’ agribusiness division invests heavily in developing new seed varieties. Due to climate change, it no longer relies on parametric distributions for yield uncertainty (output per acre). Instead, it collects test-field data and optimizes production using scenario-based stochastic linear programs with sample average approximations. Literature reports performance bounds for these programs in ambiguity spaces. The decision analysis team developed a smooth ambiguity model to balance data collection costs with improved performance, determining the optimal sample size for each seed in the portfolio. The firm successfully implemented this approach.
|