Research On Fund Investment Based On Machine Learning From The Perspective Of Market Timing And Industry Rotation

Securities investment funds have become an important part of the capital market,and its healthy development is related to the stability of the financial market and the development of the real economy.In the context of the continued growth of the fund industry,as an important investment variety,the profitability of securities investment funds should be improved for meeting the increasing investment demand of the people and the urgent expectation of high quality wealth management.There is no doubt that the market which the securities fund invests is a huge and complex evolution system.The fluctuation of asset prices is affected by various factors such as politics,economy,industry and investor sentiment.The way,degree of the influence and the frequency of price fluctuation are always different,which leads to the market changes extremely complex and difficult to control.In addition,there are nonlinear and high noise among different prices,which cause the effective investment portfolio is very difficult to be constructed.Therefore,this paper takes the main investment target of the securities fund as the analysis object,uses the theory and method of machine learning to systematically study the fund investment from the perspective of market timing and industry rotation.Firstly,this paper reviews the existing literatures and analyzes the mechanism of fund investment decision based on machine learning.Based on the definition of fund investment,this paper analyzes the basic theory of fund investment.The key links of fund portfolio management are defined from the aspects of specific stock selection,portfolio structure and fund performance evaluation.According to fund investment decision theory,and combined with the machine learning algorithms,this paper expounds the fund investment decision mechanism.From the perspective of market opportunity and industrial shift,the investment decision making and management framework of fund are established based machine learning.After that,using machine learning algorithms to explore the ability of fund to choose market timing.The choice of fund market timing is based on accurate market prediction.Based on the Gauss process,this paper constructs a market prediction model based on machine learning,discusses model structure,mean function and kernel function characteristics,design parameter estimation and model evaluation methods.At the same time,this paper takes the Chinese and US markets as the objects to carry out the relevant empirical studies.It is found that the Gauss process model is very stable for interpolation and extension prediction.In the light of this,the fund market timing selection system can accurately identify the signal and provide preconditions for the formulation of subsequent fund investment strategies.Then,machine learning algorithms are used to examine industry rotation effects and asset allocation.In this paper,the industry rotation effect is defined as the transfer process among the hidden Markov states,the industry index is the hidden state node.The industry rotation effect recognition model based on the hidden Markov method is designed.This paper selects the transaction data of 1696 listed companies from 2011 to2018,which are corresponding to 18 industries,as samples to prove that the industrial rotation effect of China's securities market industry is significant,the strong industry maintains a certain sustainability,and the probability of industry transformation is low.Based on this,the machine learning algorithms are used to estimate the industry transition probability matrix and calculate the industry asset weights.It is found that the industry transition probability matrix has a high degree of fit with the industry asset weight results.Meanwhile,it is also found that light industry manufacturing,food and beverage,medical and health,national defense and military industry,computer and other industries appear more frequently in the asset portfolio,with a large proportion.The characteristics of economic transformation and upgrading under the new normal are confirmed.Finally,according to the selection of market timing and the identification of industry rotation effect,different types of fund portfolios are constructed on the basis of preferred stocks and their performance is evaluated.The empirical results show that from the perspective of market timing and industry rotation,based on machine learning to construct the fund portfolio has achieved high performance,can adapt to the change of the market,with high performance,fitness and persistent degrees.For passive portfolios,growth portfolios,value portfolios,and balanced portfolios,investment fund has achieved higher yields,yield of different types of portfolio investment are more than120%,and the evaluation indicators based on the overall performance,performance adaptability and performance sustainability show that the machine learning portfolio has good robustness.Based on the above results,this paper proposes countermeasures and suggestions for portfolio optimization based on machine learning.