Research On Intelligent Portfolio Management And Dynamic Trading

With the continuous growth of consumption and investment demand,the investment philosophy of Chinese investors is gradually changing.Individual investors are no longer satisfied with the investment in a single asset,thus much attention is paid to portfolio management.How to promote the healthy development of multi-level capital market and effectively prevent financial risk is one of the core tasks of financial reform.This emphasizes that the capital market should play a good role in resource allocation and financing,and also makes portfolio management increasingly important in the financial market.Financial market is a complex and dynamic system,which is influenced by economic environment,investor psychology and government policy,leading to its nonlinear,unstable and complex characteristics.Portfolio management is a pretty complex and unstructured decision-making process,which involves a series of processes such as financial forecasting,investment analysis,portfolio optimisation and trading.In addition,with the development of financial big data recently,the huge amount of data puts forward higher requirements for data storage,data analysis and information technology.This situation increases the difficulty of portfolio management research.Statistics tend to model based on a large number of constraints,while early machine learning methods mainly relied on artificial feature design,which may interfere with the results.Hence,both of these methods have certain limitations in analysing complex,high-dimensional and noisy financial data.The emergence of artificial intelligence,such as deep learning and reinforcement learning,provides new solutions and ideas for these problems.Deep learning is able to identify the complex results of high-dimensional financial data effectively and transform low level features into high level and abstract features.Reinforcement learning is capable of autonomous learning,online learning and continuous decision making,and mining hidden rules from big data.It has great advantages in solving complex financial decision optimisation.In this case,artificial intelligence technologies,such as deep learning and reinforcement learning,are applied in the process of asset pre-selection,portfolio dynamic optimisation and trading.Meanwhile,based on the empirical analysis of the stock market data of China and the UK,this study puts forward the intelligent portfolio management method and dynamic trading model.Specifically,there are four main parts in this research as follows.First of all,asset preselection using deep learning.Deep learning and machine learning are applied to achieve the financial time-series prediction empirical application on big data volume.And then prediction results of different models are compared in order to take the prediction results of models with better performance as the main basis for the asset preselection.We discover the long short-term memory networks are appropriate for financial timeseries forecasting,to beat the other early machine learning models and the statistics model by a very clear margin.Second,portfolio optimisation based on deep reinforcement learning.Using the assets selected in the previous chapter,this part constructs the initial portfolio and determines the size of the portfolio.Then,the portfolio is optimised according to the asset information by taking advantage of the autonomous learning of deep reinforcement learning algorithm deep determination policy gradient in order to achieve the optimal risk-return ratio of the portfolio.We discover that,in the process of portfolio formation and optimisation,it is necessary to use machine learning for asset pre-selection.Furthermore,the long-term information of financial time series data is also quite important.Because it can help investors understand the long-term change rules of financial assets better and lay a significant foundation for the portfolio formation and optimisation.Third,portfolio optimisation considering changes in market conditions.This part is an extension of the previous section.Based on the dynamic information of external market environment,we modify the deep deterministic strategy gradient method.The modified method can adjust the learning strategy and parameter updating rules adaptively according to different market conditions,so as to adjust and optimise the asset allocation of the portfolio dynamically.It is found that in the process of dynamic portfolio optimisation,the change of external market environment is greatly crucial.Because different market conditions will not only affect the asset prices and investment returns in the portfolio,but also affect the investment behaviour of investors.Hence,it is recommended that investors not only concentrate on information about their holdings but also pay attention to market movements.Fourth,portfolio dynamic trading model incorporating investment risk preference and stop-loss mechanism.A series of portfolio dynamic trading models using deep reinforcement learning and recurrent reinforcement learning are deeply studied.In the process of trading,three objective functions containing different risk constraints are considered in order to meet different investment preferences.Besides,dynamic stop-loss is added into each trading to control the risks.The optimal trading model under different objective functions can transform the portfolio trading mode automatically according to the current market state and asset information,so as to cope with the change of different market.Moreover,through the dynamic trading between the portfolio assets and external market assets,these models can adjust the composition of portfolio assets and asset allocation in real time.It is found that when Sharpe ratio or Return rate is used as the objective function,the trading model using deep reinforcement learning method is more suitable to solve the portfolio dynamic trading.When the objective function is Calmar ratio,the trading model based on recurrent reinforcement learning method performs better.This research proposes the intelligent portfolio management method and trading model,which enriches the modern portfolio theory and financial empirical research,meanwhile,provides a reference for the research of artificial intelligence technology in the field of economic management.During the portfolio management process,this paper considers the long-term dependence of financial time series data,external market environment,investor preference as well as dynamic stop-loss mechanism successively,so as to improve and extend the portfolio optimisation and trading model.This solution provides a new perspective for research in portfolio management.