Report: Online Learning Algorithms for Portfolio Selection

This project report is published in fulfillment of a Bachelor of Science Honours degree in Statistics. Portfolio selection strategies exist which can perform as well as the best constant rebalanced portfolio and can beat the best stock in the market. In this project, three such algorithms are implemented, namely the Universal Portfolio introduced by Cover (1991), the ANTICOR algorithm introduced by Borodin et al. (2004), and the nearest neighbour based approach used by Györfi et al. (2008). Each algorithm is tested on real data from the New York Stock Exchange (NYSE). For the backtests of the ANTICOR algorithm on the NYSE data, measures of backtest overfitting introduced by Bailey et al. (2014) are calculated, including deflated Sharpe ratios and probabilities of backtest overfitting.