Mann, Andrew D;
(2022)
Machine Learning Methods to Exploit the Predictive Power of Open, High, Low, Close (OHLC) Data.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
AndrewDMannPhDFinal.pdf - Accepted Version Download (3MB) | Preview |
Abstract
Novel machine learning techniques are developed for the prediction of financial markets, with a combination of supervised, unsupervised and Bayesian optimisation machine learning methods shown able to give a predictive power rarely previously observed. A new data mining technique named Deep Candlestick Mining (DCM) is proposed that is able to discover highly predictive dataset specific candlestick patterns (arrangements of open, high, low, close (OHLC) aggregated price data structures) which significantly outperform traditional candlestick patterns. The power that OHLC features can provide is further investigated, using LSTM RNNs and XGBoost trees, in the prediction of a mid-price directional change, defined here as the mid-point between either the open and close or high and low of an OHLC bar. This target variable has been overlooked in the literature, which is surprising given the relative ease of predicting it, significantly in excess of noisier financial quantities. However, the true value of this quantity is only known upon the period's ending – i.e. it is an after-the-fact observation. To make use of and enhance the remarkable predictability of the mid-price directional change, multi-period predictions are investigated by training many LSTM RNNs (XGBoost trees being used to identify powerful OHLC input feature combinations), over different time horizons, to construct a Bayesian optimised trend prediction ensemble. This fusion of long-, medium- and short-term information results in a model capable of predicting market trend direction to greater than 70% better than random. A trading strategy is constructed to demonstrate how this predictive power can be used by exploiting an artefact of the LSTM RNN training process which allows the trading system to size and place trades in accordance with the ensemble's predictive certainty.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Machine Learning Methods to Exploit the Predictive Power of Open, High, Low, Close (OHLC) Data |
| 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. |
| UCL classification: | 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 UCL > Provost and Vice Provost Offices > UCL BEAMS UCL |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10155501 |
Archive Staff Only
 |
View Item |
