Tim Scarfe and Yuri Kalnishkan
Abstract. The paper describes an application of specialist experts techniques to prediction with side information. We pick vicinities in the side information domain to create elementary time series, use standard prediction techniques to predict for those elementary series, and then merge the predictions using specialist experts methods. Prediction with expert advice bounds ensure that optimal vicinities are selected dynamically. The algorithm is tested on the problem of predicting implied volatility of options and proves to be a viable alternative to on-line regression.
Peer reviews from ACML --https://docs.google.com/document/d/1yVa3wxlXjGLHuXzbEu3LSte75VgDdVPBNeAAv7jATIY/edit?usp=sharing