Portfolio management algorithm based on long-term prediction of assets.

Sequence data prediction is widely used in many fields. One of the most typical applications is in the financial fields, e.g., it can be used to predict the prices of assets. In this paper, we propose a new portfolio management algorithm based on long-term prediction of assets. Specially, we first predict the prices of assets for a long period through a sequence prediction algorithm which is called Informer. We then use the predicted asset price along with the trading information (e.g., the opening price, high price, low price, closing price and volume) as the environmental state of deep reinforcement learning, and obtain the trading strategies of assets through the deep reinforcement learning algorithm. To make the Informer model be more suitable for our portfolio management algorithm, we propose a new loss function, which put more focus on the trends of assets, for it. We perform experiments on both the stock market and the cryptocurrency market. The experimental results show that our algorithm can achieve higher accumulated profits and lower risk than state-of-the-art methods on both of the two markets.