Real Interest Rate Predictionusing Supervised and Unsupervised Learning Techniques
Sponsor: National Science Foundation
Researcher: Robert Michie
Advisor: Atin Basuchoudhary and John David
This project examines factors of political risk and theirability to influence and predict real interest rates. Various data miningtechniques were used on the ICRG researcher dataset for political risk, whichis made up of 12 weighted components.
The dataset contains data from the years1984 to 2008 for 37 countries. Unsupervised learning techniques were used inorder to find hidden structures within the data itself, while supervisedlearning techniques were used to make future predictions of the real interestrate.
The key points of this work arethat money growth is not a good predictor of real interest rates, as well asthe fact that a committee of artificial neural networks proved to best thestrongest model in terms of predictive performance, and finally that there arecertain political risk factors that are indicative of changes in real interestrates.