Can machine learning be used to accurately forecast intraday index returns using candlestick data?
(2025) DABN01 20251Department of Economics
Department of Statistics
- Abstract
- This thesis wants to investigate whether machine learning, trained on candlestick data, can be used to forecast intraday returns. Previous research has primarily focused on returns with longer horizons and is therefore leaving a research gap on shorter horizons. This thesis will use a combination of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) as the deep learning model to forecast the intraday returns, both as a regression task, forecasting returns, and as a classification task, forecasting return directions. Using an ARIMAX and a logistic regression model as a benchmark. Data is gathered as hourly candlestick data for the Swedish stock index (OMXSPI) from 1 January 2024 to 1 January 2025. The results indicated... (More)
- This thesis wants to investigate whether machine learning, trained on candlestick data, can be used to forecast intraday returns. Previous research has primarily focused on returns with longer horizons and is therefore leaving a research gap on shorter horizons. This thesis will use a combination of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) as the deep learning model to forecast the intraday returns, both as a regression task, forecasting returns, and as a classification task, forecasting return directions. Using an ARIMAX and a logistic regression model as a benchmark. Data is gathered as hourly candlestick data for the Swedish stock index (OMXSPI) from 1 January 2024 to 1 January 2025. The results indicated that the regression task, forecasting returns, could neither significantly outperform the ARIMAX nor accurately forecast intraday returns. However, the classification task, forecasting return directions, did significantly outperform the logistic regression as well as accurately forecast intraday returns. The findings therefore suggest that machine learning can be used to forecast intraday return directions. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9195631
- author
- Tyrstrup, Markus LU
- supervisor
- organization
- course
- DABN01 20251
- year
- 2025
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- Intraday Forecasting, Machine Learning, Candlestick data, Convolutional Neural Networks, Long Short-Term Memory
- language
- English
- id
- 9195631
- date added to LUP
- 2025-09-12 09:05:23
- date last changed
- 2025-09-12 09:05:23
@misc{9195631,
abstract = {{This thesis wants to investigate whether machine learning, trained on candlestick data, can be used to forecast intraday returns. Previous research has primarily focused on returns with longer horizons and is therefore leaving a research gap on shorter horizons. This thesis will use a combination of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) as the deep learning model to forecast the intraday returns, both as a regression task, forecasting returns, and as a classification task, forecasting return directions. Using an ARIMAX and a logistic regression model as a benchmark. Data is gathered as hourly candlestick data for the Swedish stock index (OMXSPI) from 1 January 2024 to 1 January 2025. The results indicated that the regression task, forecasting returns, could neither significantly outperform the ARIMAX nor accurately forecast intraday returns. However, the classification task, forecasting return directions, did significantly outperform the logistic regression as well as accurately forecast intraday returns. The findings therefore suggest that machine learning can be used to forecast intraday return directions.}},
author = {{Tyrstrup, Markus}},
language = {{eng}},
note = {{Student Paper}},
title = {{Can machine learning be used to accurately forecast intraday index returns using candlestick data?}},
year = {{2025}},
}