Deep Reinforcement Learning for Cryptocurrency Portfolio Allocation

Summary

This paper describes a model-less convolutional neural network that takes historical prices for a set of assets and outputs portfolio weights. It is trained through reinforcement learning, with cumulative return serving as the reward. The training data cover 0.7 years of cryptocurrency prices from an exchange, and the described trading interval is 30 minutes.

Backtests in the same market report tenfold returns over a period of 1.8 months, with results compared against several recently published portfolio-selection strategies. The excerpt does not identify the exchange, assets, benchmark details, transaction costs, or whether evaluation data were independent of training. The short training history and same-market testing limit what can be inferred about robustness or transfer to other markets; the claim that the method can apply elsewhere is not supported by results described here.

Key ideas

• A convolutional neural network maps historical asset prices to portfolio weights.
• The model is trained with reinforcement learning to maximize cumulative return.
• The reported training sample spans 0.7 years of cryptocurrency price data.
• Backtests use 30-minute trading periods in the same market and report tenfold returns over 1.8 months.
• The excerpt does not establish out-of-sample robustness or performance in other markets.

Tags

# Cryptocurrency Portfolio Management with Deep Reinforcement Learning


# Cryptocurrency Portfolio Management with Deep Reinforcement Learning









Portfolio management is the decision-making process of allocating an amount of fund into different financial investment products. Cryptocurrencies are electronic and decentralized alternatives to government-issued money, with Bitcoin as the best-known example of a cryptocurrency. This paper presents a model-less convolutional neural network with historic prices of a set of financial assets as its input, outputting portfolio weights of the set. The network is trained with 0.7 years' price data from a cryptocurrency exchange. The training is done in a reinforcement manner, maximizing the accumulative return, which is regarded as the reward function of the network. Backtest trading experiments with trading period of 30 minutes is conducted in the same market, achieving 10-fold returns in 1.8 months' periods. Some recently published portfolio selection strategies are also used to perform the same back-tests, whose results are compared with the neural network. The network is not limited to cryptocurrency, but can be applied to any other financial markets.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.