Out of the Books and Into the Woods
(2023) NEKN02 20231Department of Economics
- Abstract
- The objective of this thesis is to analyze the predictive power of the cryptocurrency limit order book for return predictions. The analysis is performed for the BTC/USD and ETH/USD directional 10-second returns, using limit order book data from three of the largest cryptocurrency spot exchanges. The analysis employs the Random forest algorithm as a classification problem. The results demonstrate that Coinbase achieves the highest average F1 scores (accuracy), followed by
Bitfinex and Gemini. When utilizing the most recent period’s features for the predictions, the F1 scores consistently exceed those of random chance, providing empirical evidence for the predictive potential of limit order book data. It is, however, observed that the... (More) - The objective of this thesis is to analyze the predictive power of the cryptocurrency limit order book for return predictions. The analysis is performed for the BTC/USD and ETH/USD directional 10-second returns, using limit order book data from three of the largest cryptocurrency spot exchanges. The analysis employs the Random forest algorithm as a classification problem. The results demonstrate that Coinbase achieves the highest average F1 scores (accuracy), followed by
Bitfinex and Gemini. When utilizing the most recent period’s features for the predictions, the F1 scores consistently exceed those of random chance, providing empirical evidence for the predictive potential of limit order book data. It is, however, observed that the accuracy varies greatly depending on the test set and the number of periods by which features are lagged. Furthermore, the study investigates the lead-lag relationships among the exchanges and the effects on predictions. Findings suggest Coinbase as the leading exchange for BTC, while indicating that Gemini was the leading exchange for ETH. However, interpreting the results for ETH is challenging due to highly imbalanced data and methodological choices. Overall, this study underscores the predictive power of limit order book data for cryptocurrency spot returns. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9119366
- author
- Molin, Elisabeth LU
- supervisor
- organization
- course
- NEKN02 20231
- year
- 2023
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- Limit Order Book, Cryptocurrency, Random forest, Market Microstructure, Lead-lag
- language
- English
- id
- 9119366
- date added to LUP
- 2023-11-24 08:57:06
- date last changed
- 2023-11-24 08:57:06
@misc{9119366, abstract = {{The objective of this thesis is to analyze the predictive power of the cryptocurrency limit order book for return predictions. The analysis is performed for the BTC/USD and ETH/USD directional 10-second returns, using limit order book data from three of the largest cryptocurrency spot exchanges. The analysis employs the Random forest algorithm as a classification problem. The results demonstrate that Coinbase achieves the highest average F1 scores (accuracy), followed by Bitfinex and Gemini. When utilizing the most recent period’s features for the predictions, the F1 scores consistently exceed those of random chance, providing empirical evidence for the predictive potential of limit order book data. It is, however, observed that the accuracy varies greatly depending on the test set and the number of periods by which features are lagged. Furthermore, the study investigates the lead-lag relationships among the exchanges and the effects on predictions. Findings suggest Coinbase as the leading exchange for BTC, while indicating that Gemini was the leading exchange for ETH. However, interpreting the results for ETH is challenging due to highly imbalanced data and methodological choices. Overall, this study underscores the predictive power of limit order book data for cryptocurrency spot returns.}}, author = {{Molin, Elisabeth}}, language = {{eng}}, note = {{Student Paper}}, title = {{Out of the Books and Into the Woods}}, year = {{2023}}, }