Crypto Proxima : An analysis of autonomous Bitcoin trading

• Main Authors: Andersson, Simon, Häggström, Jakob, Icimpaye, Urlich, Lindstedt, Hjalmar
• Format: Others
• Language: English
• Published: 2021
• Subjects: Computer Sciences; Datavetenskap (datalogi); Probability Theory and Statistics; Sannolikhetsteori och statistik
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-444804

The purpose of this project is to analyze to what extent it is possible to profit from autonomous bitcoin (BTC) trading. This was tested by using three different models in varying complexity and simulating these using real market data. The tested models were a Long Short Term Memory (LSTM) neural network, a technical analysis model based on moving averages (MA), and a random model that generates different buy and sell points. The LSTM model showed indications of possible profitability when trading at higher frequencies. For certain parameters, the model outperformed the market. The MA model was tested using a combination of different parameters for a total of 48 simulations. These had different trade offs based on the combinations, where some were able to predict the market to some extent. Regarding the random interval model, the results showed that the simulations were normally distributed around approximately half of the market gain or loss. The reason behind this is likely because the model is invested half of the time, and idle the other half. It seems that autonomous trading definitely can be profitable, however choosing the optimal parameters remains as a subject for further research.