Research On Portfolio Trading Strategy Of Chinese Stock Market Based On Temporal Network

With the development of portfolio theory,investors are increasingly paying more attention to risk in their investment and are beginning to diversify their assets.The investors care about which stocks to choose for investment,when to buy and sell the stocks,and how to allocate in each stock.The portfolio trading strategy including three steps.The first step is stock selection,which is to select stocks worth investing.It is the basis of the last two steps;the second step is market timing,different market timing the returns and risks are different;the third step is asset allocation.Non-systemic risks can be diversity through reasonable asset allocation.This thesis builds portfolio strategies around the three aspects of stock selection,market timing and asset allocation,using the Chinese stock market data as a sample and uses cross-disciplinary research methods to build a temporal stock network.A comprehensive index of temporal network is used to guide stock selection,a dynamic evaluation and adjustment system based on genetic algorithm is constructed for market timing optimization,and a information feedback temporal network is used to provide early warning signals,and a dynamic weighted asset allocation model based on particle swarm algorithm is constructed.The main research work and innovation contribution of this thesis are reflected in the following aspects:(1)In the stock selection strategy,a stock temporal network is constructed and the comprehensive index of the temporal network is proposed as the basis for stock selection.Temporal network is based on the traditional static stock network,incorporating the time attribute information,so that the topological indicators are more accurate in describing the dynamic change of the network,and the comprehensive index of the temporal network is proposed as the basis for stock selection.We test the validity and applicability of stock selection method by comparing the result of out-of-sample test with static network indicators,benchmark indicators and financial market common indicators under different market environments.In the modeling of temporal networks,considering the nonlinear characteristics of stock price fluctuations,the mutual information of the entropy theory is used as a measure of the correlation between stocks,and the minimum spanning tree is used to retain the most critical information in each time layer.After the out-of-sample inspection,it is found that choosing the center stocks of temporal network can outperform the market benchmark indicators and most commonly used indicators in most financial markets in the short-term,but when the long-term the indicators fail,the comprehensive indicator have temporal effect.(2)In the stock market timing strategy,we constructed a market timing model based on temporal network portfolio selection.Comprehensive use of genetic algorithms and early warning signals of information feedback temporal network to provide decision support for stock trading opportunities.Faced with the risk of severe market fluctuations and market downtrend,a market timing model based on the temporal network portfolio selection is constructed aiming at the market timing of buying and selling.From two perspectives,this thesis first uses genetic algorithm to optimize the technical indicators of adaptive moving average as a signal of market timing,secondly,we construct information feedback temporal stock and uses the early warning indicator as a prohibition on buying to prevent market risks.Through the outof-sample testing,it provides investors with decision support in different market conditions.(3)In the asset allocation strategy,a dynamic asset allocation optimization model based on the temporal network portfolio selection is constructed to provide the optimal asset allocation weights.Fixed asset allocation weights are used in the research on the existing network stock selection,using particle swarm algorithm to dynamically evaluate and continuously adjust the asset allocation weights during the training period,and give the optimal asset allocation before the start of the test period.By comparing with the asset allocation methods used in existing research,it is verified in different market environments the effectiveness of the asset allocation strategy,combined with the industry of the stock,we provide support for asset allocation investment decisions in different market conditions.