A carregar...
Publicação
Trend Following Algorithmic Trading with a Reduced Universe of Cryptocurrencies: Using unsupervised learning for universe reduction and trend following strategies to generate profits
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.43 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Cryptocurrencies have revolutionized trading with their decentralized and volatile
nature, presenting unique opportunities for trend-following strategies. This study
investigates the enhancement of such strategies by implementing a reduction in the
universe of cryptocurrencies based on their similarities, aiming to improve trading
performance. Additionally, a market momentum classification will be constructed to
analyze and compare performance across different market momentum environments.
By utilizing Support Vector Machines (SVM) and Principal Component Analysis (PCA), the
overall universe of cryptocurrencies is reduced to the most similar cryptocurrencies
based on metrics of momentum, growth, and technical analysis. This approach aims to
address market overload caused by the considerable number of available
cryptocurrencies, removing the most erratic cryptocurrencies.
To assess the efficacy of two trend-following trading strategies—an equally weighted
portfolio approach and a time-weighted portfolio approach—we will utilize a
momentum-based trading strategy serving as our benchmark. All strategies will be used
with the reduced universes produced by our method. Additionally, a market momentum
classification will be constructed using market capitalization data of the crypto market,
with the goal of gaining relevant insights into the effects of the overall market
momentum environment on the performance of the algorithmic trading strategies and
reduced universe combinations.
The innovation lies in the method of reducing the cryptocurrency universe by identifying
the most similar tokens during the same period while asserting the value of
unsupervised learning models like SVM and PCA in enhancing trend-following trading
strategies by reducing complexity.
In summary, this research underscores the critical differences in short and long trend
formation periods, the effectiveness of trend-following approaches, and the superior
performance of time-weighted portfolios, demonstrating how SVM and PCA can
enhance algorithmic trading strategies by reducing complexity and optimizing
profitability in the dynamic cryptocurrency market.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Crypto Currencies Machine Learning Unsupervised Learning Algorithmic Trading Trend Following SDG 9 - Industry, innovation and infrastructure