The popularity of cryptocurrencies exploded in 2017 due to several consecutive months of super exponential growth in their market capitalization , which peaked at over $ 800 billion in January 2018. Today, there are more cryptocurrencies out there. are actively traded. According to a recent survey , between and millions of private and institutional investors are present in the various transaction networks and market access has become easier over time. Major cryptocurrencies can be purchased using fiat currency on a number of online exchanges (e.g. Binance , Upbit , Kraken , etc.) and can then be used in turn to buy fewer cryptocurrencies. popular. The daily trading volume is currently over $ 15 billion. Since 2017, more than 170 specialized cryptocurrency hedge funds have emerged and Bitcoin futures have been launched to meet the institutional demand for Bitcoin trading and hedging .
The market is diversified and offers investors many different products. To name a few, Bitcoin was expressly conceived as a medium of exchange [7, 8]; Dash offers advanced services beyond the Bitcoin feature set, including instant and private transactions ; Ethereum is a blockchain-based public distributed IT platform with smart contract (scripting) functionality, and Ether is a cryptocurrency whose blockchain is generated by the Ethereum platform ; Ripple is a Ripple Real-Time Gross Settlement (LBTR), currency exchange and remittance network , and IOTA focuses on providing secure communications and payments between agents on the Internet of Things .
The emergence of a self-organized market for virtual currencies and / or assets whose value is generated primarily by social consensus  has naturally attracted the interest of the scientific community [8, 14-30]. Recent results have shown that the long-term holdings of the reported cryptocurrency remained stable between 2013 and 2017 and are compatible with a scenario where investors simply sample the market and allocate their money based on the cryptocurrency’s market shares [ 1]. While this is true on average, several studies have focused on analyzing and predicting price fluctuations, mainly using traditional approaches for analyzing and forecasting financial markets [31-35].
The success of machine learning techniques for stock market forecasting [36-42] suggests that these methods may also be effective in predicting cryptocurrency prices. However, the application of machine learning algorithms to the cryptocurrency market has so far been limited to the analysis of Bitcoin prices, using random forests , Bayesian neural network , short-term memory neural network for algorithms a long term  and others [32, 46]. These studies were able to anticipate Bitcoin price fluctuations to varying degrees and revealed that the best results were achieved using algorithms based on neural networks. Deep reinforcement learning has been shown to outperform uniform buy and hold strategy  by predicting the prices of 12 cryptocurrencies over a one-year period .
Other attempts to use machine learning to predict the prices of cryptocurrencies other than Bitcoin come from non-academic sources [49-54]. Most of these analyzes focused on a limited number of currencies and provided no benchmark comparisons for their results.
Here, we test the performance of three models to predict the daily price of cryptocurrencies for 1,681 coins. Two of the models are based on increasing gradient decision trees  and one is based on recurrent neural networks with short-term long-term memory (LSTM) . In all cases, we build forecast-based investment portfolios and compare their performance in terms of return on investment. We found that all three models work better than a “simple moving average” reference model [57-60] in which the price of a currency is predicted as the average price of the previous days and that the short-term memory method Neural networks consistently produce the best return on investment.