Estimating the impact of weather on CBOT corn futures prices using machine learning

Abstract

I apply machine-learning methods to study the impact of hourly changes in county-level weather in major corn-producing US states on the Chicago Board of Trade (CBOT) corn futures prices. Futures prices respond to shocks in expected production levels, which in turn depend on weather outcomes. The percentage change in futures prices at daily/weekly/weekend frequency is forecasted using a convolutional neural network that exploits the spatial characteristic of the data. Additionally, the direction of changes in futures prices is predicted using a suite of classification models viz. logistic regression, support vector machine, and decision trees. Analytically, I compare the outcome of machine learning models on big data and test the effectiveness of high-frequency weather data in predicting futures prices. The results suggest that weather forecast data provides important information that efficiently moves the market.