UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

Online learning in financial time series

Borrageiro, Gabriel Franceisco; (2023) Online learning in financial time series. Doctoral thesis (Ph.D), UCL (University College London). Green open access

[thumbnail of Gabriel_Borrageiro_PhD_Thesis.pdf]
Gabriel_Borrageiro_PhD_Thesis.pdf - Published Version

Download (2MB) | Preview


We wish to understand if additional learning forms can be combined with sequential optimisation to provide superior benefit over batch learning in various tasks operating in financial time series. In chapter 4, Online learning with radial basis function networks, we provide multi-horizon forecasts on the returns of financial time series. Our sequentially optimised radial basis function network (RBFNet) outperforms a random-walk baseline and several powerful supervised learners. Our RBFNets naturally measure the similarity between test samples and prototypes that capture the characteristics of the feature space. In chapter 5, Reinforcement learning for systematic FX trading, we perform feature representation transfer from an RBFNet to a direct, recurrent reinforcement learning (DRL) agent. Earlier academic work saw mixed results. We use better features, second-order optimisation methods and adapt our model parameters sequentially. As a result, our DRL agents cope better with statistical changes to the data distribution, achieving higher risk-adjusted returns than a funding and a momentum baseline. In chapter 6, The recurrent reinforcement learning crypto agent, we construct a digital assets trading agent that performs feature space representation transfer from an echo state network to a DRL agent. The agent learns to trade the XBTUSD perpetual swap contract on BitMEX. Our meta-model can process data as a stream and learn sequentially; this helps it cope with the nonstationary environment. In chapter 7, Sequential asset ranking in nonstationary time series, we create an online learning long/short portfolio selection algorithm that can detect the best and worst performing portfolio constituents that change over time; in particular, we successfully handle the higher transaction costs associated with using daily-sampled data, and achieve higher total and risk-adjusted returns than the long-only holding of the S&P 500 index with hindsight.

Type: Thesis (Doctoral)
Qualification: Ph.D
Title: Online learning in financial time series
Open access status: An open access version is available from UCL Discovery
Language: English
Additional information: Copyright © The Author 2022. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author’s request.
Keywords: online learning, transfer learning, reinforcement learning, prediction with expert advice, financial time series
UCL classification: UCL
UCL > Provost and Vice Provost Offices > UCL BEAMS
UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science
UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Computer Science
URI: https://discovery.ucl.ac.uk/id/eprint/10163818
Downloads since deposit
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item