Exploration of ethereum investment strategy with HMM

• Main Authors: Ng, Jacqueline Jia Yee, Ng, Yongwen, Wong, Wei Hao
• Other Authors: Low Chan Kee
• Format: Final Year Project
• Language: English
• Published: 2018
• Subjects: DRNTU::Humanities
• Online Access: http://hdl.handle.net/10356/73603
• Institution: Nanyang Technological University

In this paper, we provided a general insight into the burgeoning cryptocurrency market. Having inspected the capabilities of cryptocurrencies as investment assets, we saw value in developing a price prediction model in this highly volatile market. Ether was chosen as the subject due to its unparalleled potential amongst its competitors. Driven by the similarities presented between cryptocurrency and stock, coupled with evidence of hidden states in the financial market, this paper conducted the first application of the Hidden Markov Model to the cryptocurrency space. Our main objectives were to first model the time series data of Ether since its establishment, then use the trained model to forecast future closing prices of Ether before finally devising an investment strategy. HMM was used to solve three fundamental problems namely the Evaluation Problem, Learning Problem and Decoding Problem. Given 936 observations of daily Ether prices obtained from Yahoo Finance, 80% of this data was used as a training set while the remaining 20% was used as the testing set. The Forward algorithm and Baum-Welch algorithm helped obtained the model parameters. Thereafter, Viterbi algorithm decoded the likely state sequence of the observations. Using Mean Absolute Percentage Error as the indication of forecasting power, the selected Hidden Markov Model (HMM) has 3 states and 3 mixtures of Gaussian distribution, with the lowest MAPE of 4.63568. Via Monte Carlo simulations, our HMM investment strategy produced a superior weekly return of 5.68% as compared to 3.94% for the “naïve” strategy. This indicated the successful adaptation of HMM in the cryptocurrency market despite limitations from the inherent assumption. Future extensions to the paper could include the use of more model inputs, account for transaction fees and consider the heavy correlation between cryptocurrencies. We also anticipate that our findings require re-validation over time given the rapidly evolving nature of the market.