Hedge Funds Selection

In recent years,the hedge fund industry has been developing vigorously,especially since 2015,the overall scale has been rapidly expanded.Because of the non-standard disclosure of information and the more difficulty of data acquisition,there are relatively few studies on hedge funds.With the rapid development of hedge fund market,the investment strategies are gradually diversified.The characteristics of hedge funds with different strategies are different,which brings certain challenges for investors to select funds.Under this background,the study of the selection and prediction of the return rate of hedge funds can help investors to build up their investment confidence,provide a basis for investors to select hedge funds better and a reference for regulatory agencies to formulate regulatory policies.In the current research,most of them focus on the construction of factors to evaluate performance of the hedge funds,or through the combination of these factors to form a comprehensive rating of hedge funds,and these factors are based on the historical performance of funds.However,have these risk factors based on historical data any influence on funds' future return? Persistent influence is an important issue that needs to be studied.If the risk factors based on historical data do not have a lasting influence on the future performance,investors can not make choices on hedge funds based on historical evaluation factors.Therefore,whether the evaluation factors based on historical performance can predict the future return of hedge funds is an important issue we are concerned about.Because the information disclosure of hedge funds is limited,this paper constructed risk factors describing the return characteristics of hedge funds based on the available return information,to study whether the risk factor describing the return characteristics of hedge funds has predictability to its future return.Because the method of machine learning does not emphasize the structure of the model and does not need prior statistical assumptions,this paper constructed a supervised machine learning model based on the multi-factor model.According to the prediction ranking results,the hedge funds are divided into ten equal-weight portfolios,and the prediction model is assessed by observing the ranking of the real returns of the portfolio.In this paper,linear,non-linear,model shrinkage and model average machine learning methods are used to reduce the possible errors caused by model setting.Because the hedge funds with different strategies have different investment targets and basis,this paper makes respectively an analysis of different strategies of hedge funds.For the same strategy of hedge funds,different machine learning methods are combined separately with three months,six months and twelve months of prediction window period to simulate the closure period of hedge funds,and the macro-factor prediction model proposed by Avramov,Barras and Kosowski(2013)is used as the benchmark comparison model.We select the portfolio of hedge funds,and compare the performance of the portfolio with the performance of the market index of the corresponding strategies based on the predicted results.Finally,the contribution of risk factors is analyzed by random forest machine learning method,And the relationship between the return time series of fund portfolio and main market indexes is analyzed to further verify the validity of selecting fund portfolio based on the prediction results.The main conclusions of this paper are as follows:(1)For the hedge funds with specific strategies,the risk factors describing their return characteristics can predict their future cumulative return.(2)The risk factors describing the return characteristics of hedge funds with different strategies need to combine different predictable window periods to predict the funds' future cumulative return.(3)Among the risk factors describing the return characteristics of hedge funds,the return over the past period contribute more to funds' future cumulative return.(4)Machine learning methods with model shrinkage,model averaging and non-linear calculation of factors have better prediction effect,and the fund portfolio selected performs better.(5)To a certain extent,prediction results based on return characteristic factors can guide investment to a certain extent.The excess return of the selected portfolio comes from the model itself rather than the market.The main innovations of this paper include:(1)Analyze and describe the predictability of the return ratio factor of hedge fund to its future cumulative return which can better guide investment.(2)Hedge funds with different investment strategies are studied and analyzed,and the results are more detailed and accurate.(3)Using a variety of machine learning methods to train the datas of hedge funds to reduce the impact of model setting errors.The inadequacies of the paper are as follows:(1)The paper mainly focuses on predicting the future cumulative return of the fund without considering its risk.(2)Hedge funds are ranked according to the forecast results of return rate,without considering the specific value of return and the overall performance of the market.(3)Because of the difficulty in obtaining information of hedge funds,only the risk factors describing the return characteristics of funds are selected for analysis,which is not all-round enough.