A carregar...
Publicação
Cryptocurrency portfolio optimization using genetic algorithms
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.54 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Blockchain most discussed application has been in cryptocurrency, being
Bitcoin its first. Unbeknownst to many, Bitcoin not only introduced a new
digital means of exchange creating fertile ground for others trying to emulate
it. Cryptocurrencies took relevance beyond computer science spheres to reach a
place of relevance as a security for several investors, making it relevant to study
beyond computer science spheres. Economists, statisticians, and portfolio managers,
have taken the subject of cryptocurrencies as case study for open digital
finance, opening questions regarding price behaviors and objective investment
strategies, creating opportunities of research on fields such as machine learning.
Nevertheless, each cryptocurrency has its technicalities, and different value
proposals, making the subject relatable, in some sense, to traditional financial
instruments, from which some lessons can be of use, for example, how it is difficult
to come up with a unique approach or toolset that forecasts prices and
optimizes investments, at least in a general sense. Here the objective is to tackle
a cryptocurrency portfolio optimization by means of genetic algorithms. Two
approaches are suggested, the first Limited Trading approach, consists on using
genetic algorithms to find how much of the coins to invest for profit considering
to sell at the last day. Second approach is called Open Trading, much like the
first, intends to find profit but it allows buying and selling at each timestep. In
both cases respecting budgeting limitations and using forecast price values of
each coin, but for this, instead of delving into the complexities of thousands of
possible coins and algorithms, it was used a combination of machine learning
methods to forecast prices of the coins in the portfolio. It was found that Limited
trading outperforms its counterpart in fitness (expected portfolio value) and
%Return. It was also found a genetic algorithm parametrization successful for
both strategies, and highlighting the value of the theoretical proposal for fitness
optimization based on matrix operations for Open Trading, which has room for
improvement and further development in fields beyond portfolio optimization.
Descrição
Project Work presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Blockchain