Portfolio Algorithm Based On Long-Term Asset Price Prediction

Financial market is the key component of national capital market,which plays a vital role in the rapid development of national economy.However,financial asset prices are characterized by fast changing speed,many interfering factors and insufficient periodicity,so it is difficult to accurately predict them by using traditional analysis methods.With the rapid development of artificial intelligence,machine learning and deep learning are widely used in asset price forecasting,which brings great opportunities for investors.Asset price prediction can help investors to buy and sell assets at the most appropriate time.The portfolio problem is exactly the question of whether to buy or sell an asset at each point in time in order to maximize investment returns and minimize risk.However,portfolio problems are a long-term decision-making process,which means that if this paper can predict asset prices over a period of time in the future,investors can make more favorable decisions.Portfolio asset trading is a game process with incomplete information and it is difficult for the single objective supervised learning model to deal with such serialized decisionmaking problems.Reinforcement learning is a method of learning by interacting with the environment and gradually improving its performance through trial and error.Therefore,reinforcement learning is widely used in asset management problems.This paper explores the potential of reinforcement learning in asset trading strategies to maximize portfolio returns.The main work of this paper includes the following two aspects:First,this paper proposes a new portfolio algorithm LTPA based on long-term asset prices.Firstly,asset prices are forecast over a long period of time by a long-term sequence data prediction algorithm.Then,using the predicted asset prices and trading information(such as the opening price,the highest price,the lowest price,the closing price and the trading volume)as the environment state of deep reinforcement learning,the trading strategy of the asset is obtained through the deep reinforcement learning algorithm.In order to make the long-term sequence data prediction model more suitable for the portfolio algorithm in this paper,a new loss function is proposed which pays more attention to the forecast trend of assets.Numerous experiments have been conducted in the stock market and cryptocurrency market.The experimental results show that the proposed algorithm can obtain higher cumulative profit and lower risk than the existing methods in both markets.Secondly,this paper proposes a portfolio algorithm ECPA based on expected asset price coding,which is capable of making end-to-end portfolio decision.In this paper,technical indicators are introduced on the basis of asset trading information,so as to improve the accuracy of the long sequence data prediction model.Then,the strategy function of reinforcement learning is constructed by using the coding method based on the long sequence data prediction model.In order to make the strategy function make more favorable decisions for investors,this paper introduces short-term sales into the strategy function,sells the assets predicted to decline,and then reinvests the liquidated funds in the assets predicted to rise in price,so as to improve the return of the portfolio.Experiments on real stock data sets show that the algorithm can enable investors to obtain higher returns at the lowest possible risk.