by Laura Kahn
Algorithmic Trading of Coffee Futures with Machine Learning
Abstract: Data science regression techniques are applied to predict financial commodities prices. Machine learning algorithms are used to predict the daily closing price of coffee futures with a maximum percent prediction error of 0.00328%.
Data: The final data set consisted of 2,805 daily observations of coffee futures prices from January 1, 2010 - November 15, 2017. Three input features are : open, high and low price in US Dollars. Target feature is closing price in US Dollars.
Results: Linear Regression goodness of fit of 0.0996 and prediction error of 0.00009736. Decision Tree Regression goodness of fit of 0.799 and prediction error of 0.00000542. Decision Tree Regression with AdaBoost goodness of fit = 0.749 and prediction error of 0.00328. Ridge Regression goodness of fit of 0.996 and prediction error of 0.00009738.