Title: Aggregating strategies for online portfolio optimization
Authors: Christine Keribin - INRIA - Paris-Saclay University (France) [presenting]
Thomas Prochwicz - INRIA - Universite Paris Sud (France)
Abstract: Online portfolio selection sequentially selects a portfolio over a set of assets, aiming to maximize the cumulative wealth. There is a large number of selection algorithms, that can be classified according to their leading principle: buy and hold, follow-the-winner, follow-the-user or pattern-matching approaches for example. The performance of these algorithms can depend on the market trend, some being more adapted in declining or in upraising period. These strategies can be viewed as experts, and they can be aggregated with online computed weights to propose a predictor than can be guessed better than the best expert. In this communication, we first compare standard strategies with a strategy based on a machine learning prediction. Then, we introduce aggregating algorithms and discuss their calibration. We compare the performance of the aggregation with these of the experts, and with prescient strategies. We examine the number of aggregated experts and the influence of the market trend. All the experiments are based on the CAC40 asset prices.