Aluno: Wenyan Zhao
Resumo
This study explored the impact of different indicator factor models on the
performance of portfolios under the same trading strategy within the same time frame. The indicator factors selected were the Commodity Channel Index (CCI), Volume, and Bollinger bands (Boll count). The trading strategy employed is known as the
'Long-short pairing strategy', which involves taking a long and holding position in a
single coin ranked first while short-holding a coin ranked last at the same time. 264
Coins that are traded on Binance plarforms are included in this research, and coin
selection process is refreshed every 6 hours. The time frame considered for this
analysis spans from September 20, 2020, to September 20, 2023. The evaluation
metrics for portfolio performance are the accumulated net value, annual return, portfolio standard deviation, Sharpe ratio, maximum draw-down, information ratio, etc. The hyper-parameter of the factors was tuned by machine learning using Grid
search. The results showed that combining additional factors into the model can increase
annual returns. However, reducing portfolio volatility and risk exposure is in doubt. Also, the impact of additional factors on portfolio performance and risk can vary
significantly depending on the specific factors and their correlations, emphasising the
importance of careful factor selection and combination in
Trabalho final de Mestrado