We showed a possible approach to predicting the movement of prices
of cryptocurrencies. We used a data set of four trading pairs USDT BTC,
USDT LTC, USDT ETH, and USDT XRP, gathered through the public api
of the cryptocurrency exchange Poloniex. The translation of the problem
to three possible trading actions buy, sell and hold as well as an advanced
technical analysis in combination with advanced neural net architectures was
shown as successful with a better final outcome and lower standard deviation
than just buying and holding the currency. Apart from this one of the main
contributions is a comparison of different trading strategies in combination
with the usage of neural nets. A comparison was made of three different
trading strategies. These are interval trading, trend trading and trading
with only a part of the assets. An approach of finding the minimum and
maximum values in a given 24 hour interval, called interval trading, was
shown as the most successful. We also introduce a modular framework that
was implemented during research and can be used as a quick way to check
different strategies and approaches. We used recurrent and long-short term
memory neural networks.
|
