Performance Prediction Model Of Private Equity Fund Based On Machine Learning

In recent years,to stimulate social vitality and promote economic growth,the government has actively called for “mass entrepreneurship and innovation” and a number of guidelines and policies has been introduced to set up a platform for it.The rise of “innovation and entrepreneurship” has also greatly promoted the development of China's private equity investment market.Private equity has been developed in China for more than 30 years.Its investment activities are generally carried out through the establishment of a dedicated private equity fund.In recent years,China's private equity investment market has been continuously improved,and some new features of PE fund have emerged,such as: the scale of private equity fund is growing rapidly,and its influence is gradually expanding;the quantity of local capital investment exceeds the non-local capital investment;the investment stage is concentrated in the maturity and expansion;the investment area shows the trend of decentralization;the investment industry is concentrated in high-tech,high value-added industries.All in all,more and more investors and entrepreneurs attach importance to China's private equity investment,which has become an important force in China's financial market.In this context,it is particularly important to conduct a detailed study of the performance of China's private equity fund.Because of the confidentiality of private equity,China's private equity funds do not disclose their investment performance regularly.Its performance is measured by the return of investment project when it exits;and in the Chinese private equity market,there are very few funds with two or more successful investment exit experience,so the concept of private equity fund performance is basically the same as that of private equity investment project return.In this paper,the research on the performance of private equity funds is based on the data of investment project level.Based on theoretical analysis and literature review,this paper obtains 2776 sets of private equity investment data from January 1,2009 to December 31,2017,and chooses relevant explanatory variables including project characteristics,investment plans,institutional characteristics and macro-environment.On this basis,this paper studies the performance of China's private equity fund from three aspects:Firstly,this paper uses linear regression model and one-way ANOVA to study the influencing factors of private equity fund performance.Secondly,based on the analysis of influencing factors,this paper uses a variety of machine learning methods to establish regression prediction model,and carries out the analysis of variable importance and partial dependence.Finally,a variety of machine learning methods are used to establish a private equity fund performance classification prediction model and it can provide investors with valuable investment advice.This paper mainly draws the following four conclusions:Firstly,the performance of private equity fund in China is obviously better than that of public market investment,but in recent years,the profit advantage of private equity investment market is shrinking and the investment risk is increasing,so investors should be cautious in investing.Secondly,on the whole,the explanatory variables related to investment plan,macro-environment and project characteristics have a greater impact on the performance of private equity fund,while the variables related to institutional characteristics have a smaller impact.Different regions and stages of investment have different influencing factors on the performance of private equity fund.In order to achieve higher investment returns,investors should analyze the specific situation of investment projects and attach more importance to the relevant factors.Thirdly,the regression model of private equity fund performance based on machine learning has been greatly improved compared with traditional methods,and the explanatory power of random forest model has reached 58.5%.But,generally speaking,the prediction result of the model is poor,which indicates that it is difficult to predict the performance of private equity fund directly on the basis of existing data.Fourthly,the random forest classification model has good performance in predicting the performance level of private equity fund.The overall accuracy of the model reaches 81.59%,which shows that the classification results of the model are credible.Specifically,the prediction accuracy of the model for high,medium and low returns is 77.21%,92.19% and 54.05%.From the confusion matrix of the model,it can be seen that the classification error of the model mostly occurs between adjacent categories,that is,the influence of the classification error can be controlled,which ensures the scientificity of the model.Compared with the previous literature,this paper mainly has the following two innovations:Firstly,innovation of research content.From project characteristics,investment plans,institutional characteristics and macro-environment,this paper chooses relatively comprehensive influencing factors indicators,and studies the impact of main explanatory variables on private equity fund performance from both linear and non-linear perspectives.Besides,on the basis of the research on influencing factors of private equity fund performance,this paper uses a variety of machines learning methods,establishes the regression prediction and classification prediction model of private equity fund performance,enriches the research content in the field of private equity fund,and provids operational suggestions for investors to choose investment projects and design investment plans.Secondly,innovation of research methods.The existing literature mainly studies the performance of private equity fund based on linear regression model.This paper introduces one-way analysis of variance and multiple machine learning algorithms to reflect private equity fund from different angles,and establishes a performance prediction model with good prediction performance.