Research On Asset Allocation Scheme Based On OGARCH Model And Recurrent Neural Network Model

As the most important part of the financial market,the stock market has always been inextricably linked with people's lives.Since the establishment of the Chinese stock market,investors have never stopped studying how to obtain excess returns in the market.In the past 30 years,the Chinese stock market has fluctuated sharply,and investors have gradually paid attention to the necessity of risk management.Portfolio investment is a way of risk diversification,but in the process of building a portfolio,how to choose stocks and how to allocate funds in them became a concern.The Markowitz portfolio theory provided a theoretical basis for asset allocation.However,the assumptions of MV model are strict,and the input parameters in the model are difficult to quantitatively estimate,so the practicability of the model is not strong.In order to study the covariance estimation between multiple assets,Bollerslev(1988)proposed early multivariate GARCH model to describe the volatility spillover effect of multidimensional variables,and provide model support for the covariance estimation between assets.However,because the model can not overcome both two problems of “dimensionality disaster” and dynamic correlation at the same time,the covariance matrix based on multivariate GARCH estimation is not practical in Markowitz portfolio theory.Alexander(2001)proposed OGARCH model,which uses principal component analysis to extract orthogonal factors,and then uses the GARCH model to predict the risk of each orthogonal factor.It studies the covariance estimates of large portfolios to characterize dynamic correlation between assets in the portfolio.Based on these models,the research on portfolios is very extensive,however,at the same time,a large number of scholars are keen to analyze the short-term trend of individual stocks.The recurrent neural network model(RNN)proposed by Jordan(1986)considers the correlation between sequences dynamically,which is exactly in line with the characteristics of financial assets.So it is gradually applied to the prediction of financial assets.The long-short-term memory model(LSTM)proposed by Hochreiter et al.(1997)adds memory storage units based on RNN to make the model have better memory ability..With the increasing emphasis on risk diversification,more and more investors are diversifying their assets.There are two issues when building a portfolio in the stock market: First,which industries should be selected.Second,how much funds should be allocated in each industry.Based on Markowitz's portfolio theory,this paper proposes an asset allocation scheme to solve these problems and dynamically manage portfolios.In addition,OGARCH model is introduced to predict the covariance and put it into the MV-model to construct a dynamic management scheme.We selects the week-yield of the A-share industry index as the research object,and construct medium-term portfolio to achieve weekly management.Further,investors often want to take advantage of the daily fluctuations of stocks and adjust the key stocks within weeks to compensate for the shortcomings of weekly trading.Therefore,we use LSTM model to predict stock price,and formulate short-term trading plan,which is an optimization and expansion of medium-term portfolio management.The empirical analysis shows the following conclusions:(1)The prediction ability of the OGARCH model on variance and covariance exceeds the GARCH and CCC-GARCH models respectively.(2)the back-test performance of portfolio based on OGARCH in the bull market,bear market and shock market all exceeded CCC-GARCH,and outperformed the Shanghai composite index,indicating that this scheme based on OGARCH is effective.(3)the scheme suggests that investors can appropriately improve risk tolerance in the bull market,and in other markets,risk should be minimized as investment premise.(4)for investors who are extremely risk-averse,it is not recommended to diversify investment in markets where there is no obvious profit.(5)LSTM model is effective to make short-term adjustments to stocks in the stock pool.The main innovations in this paper are as follows:(1)based on the model's advantages,the positive definiteness of the covariance matrix can be guaranteed,and the dynamic correlation between assets can be characterized while avoiding the dimension disaster.(2)we use the LSTM model to make daily adjustments to individual stocks,and better grasp the fluctuations of stocks within a week,which is optimization and expansion of the medium-term allocation scheme.(3)the program consider about investors with different risk tolerance on different market circumstance to give the corresponding selection recommendations and adjustments,providing a more comprehensive reference for investors in the real stock market investment.Due to the limited research level,there are still some shortcomings in this study:(1)the impact of stock transaction fees,taxes and other factors on the backtest results is not considered.(2)only GARCH and CCC-GARCH were selected as the comparison model,and no more complex models were provided to compare with OGARCH.(3)only the short-term adjustment of BOE is discussed,so the scope of the study is too small to judge whether the LSTM model can be widely applied to the Chinese stock market.Therefore,further research can complement more complex models and further optimize the OGARCH model by combining their respective advantages.At the same time,more individual stocks can be selected for research,and the short-term adjustment of individual stocks can be fully combined with the medium-term portfolio scheme to further improve the investment scheme.