|Ph.D Student||Uziel Guy|
|Subject||Leveraging Mechine Learning Algorithms in Online Portfolio|
|Department||Department of Computer Science||Supervisor||Professor Ran El-Yaniv|
|Full Thesis text|
Online portfolio selection, aiming to optimize the allocation of wealth across a set of assets, is a fundamental research problem in computational finance and machine learning. Aside from the theoretical challenges, the implementation of a real-world trading system is also extremely difficult. The field has been extensively studied across several research communities, including finance, statistics, coding and information theory and machine learning. Today, many aspects of it still remain unresolved. We begin this dissertation with a discussion of the problem of online portfolio selection in the presence of transaction costs. To address this issue we propose two algorithms that will enable a trader to enhance any existing commission-oblivious algorithms. The first algorithm is based on a novel regularization method. This regularization forces the trader to put more weight on a previously selected portfolio, thus reducing the transaction costs during the trading. The second algorithm is based on the observation that to effectively handle commissions, a dilution in the number of trades is necessary. This observation is exploited by using a novel mechanism that combines the predictions of long-and short-term experts, leading to an effective method to handle commissions.
We then discuss an approach to handle the risk incurred while trading. First, we review multi-objective online learning, where we propose a novel algorithm to address this problem in case when the underlying process is stationary and ergodic. We prove that under mild conditions, our algorithm is universal and thus asymptotically achieves the best possible outcome in hindsight. Later on, we show how this method can be utilized in order to incorporate the well-known risk proxy, conditional value at risk (CVaR), in online portfolio selection. Finally, we deal with the pattern matching algorithms, and propose a novel approach to incorporate the learning of a suitable kernel using a deep neural network.