Empirical Research On FOF Investment Based On Ensemble Methods And Risk Parity

With the gradual opening up of the domestic capital market,residents’ demand for value-added investment is also increasing,and investment options and financial instruments have become abundant.However,there are not many investment products that can really bring good returns to investors.To resolve this contradiction between supply and demand well,it is bound to form an institutionalized trend for investors in the future.In China,as a major institutional investor,public funds have seen a substantial increase in the number of public funds that have been established in recent years.However,most funds have not consistently achieved excess returns that exceed the market.This is also a challenge for individual investors to choose Reliable funds.At the same time,the scale of domestic private equity funds has also seen rapid growth.As of the end of last year,the management of private equity funds has exceeded 11 trillion.Under the trend of a sharp increase in incremental funds,this type of investment in the FOF Fund has begun to receive attention and favor.With the approval of the public fund FOF,domestic scholars and the industry have conducted a series of researches on FOF investment.They mainly study the common strategies of mature capital markets,such as lifecycle strategy,arbitrage strategy,risk parity strategy,fund selection strategy,etc.An empirical analysis of the applicability of these strategies in the domestic market.The construction of the FOF portfolio is generally divided into two steps,the asset screening phase and the weight configuration phase.On the basis of the existing research,this paper uses stocks,QDII,hybrid,currency,and bond funds as alternative funds,and introduces an integrated approach to machine learning algorithms as a method for constructing funds for the first phase of FOF investment screening.The weighted allocation of these five types of funds was combined with the risk parity model.Finally,the empirical analysis of the combination of these two models can bring about continuous and steady gains.The ensemble method is a machine learning algorithm that can combine multiple base learners(such as decision tree,SVM,K nearest neighbors,etc.).This algorithm is also a good performing,widely used algorithm in building prediction models.In this paper,a gradient boosting algorithm and a random forest algorithm based on a binary decision tree are used to analyze and compare the learning performance of the two algorithms,and the better learning results of the two are used as the attribute variable weights for further screening the fund.The basis for the fund was eventually obtained from the fund targets in the various positions in the back-test interval.Then use the four models commonly used in asset allocation theory,namely common risk parity model,leveraged risk parity model,minimum variance model and equal weight model to weight the asset portfolio.Finally,the performance of the four models is evaluated through four performance indicators.The empirical analysis results show that the use of ensemble methods for fund selection can obtain fund varieties with large performance potential,and without considering the relevant capital and time costs,the use of risk parity model to configure weights can achieve higher risk-adjusted investment returns under the premise of controlling overall risks.It provides new conclusions and new directions for FOF investment research.