Asset Allocation Based On The High-frequency Financial Data In The Presence Of Noise And Jumps

With the continuous development of economic globalization,countries have become increasingly economically connected.The exchange of goods and services has been increasing,bringing rich commodity resources to countries and providing developing countries with opportunities for development.However,the linkage of global economic and financial relations caused by economic globalization has made the world economy more fragile.The economic and financial crises within each country have spread abroad to other countries.Recent examples include the sovereign crisis of European countries and the global financial crisis in 2008.The 1997 Asian financial crisis,and further examples are the oil crisis of the 1970 s.In the context of economic globalization and frequent financial crises,it is very important to study the causes of crisis formation and to identify and respond to the crisis.In general,the outbreak of the crisis is due to the collective release of asset price risk to a certain extent.Therefore,we can identify the occurrence of the crisis and judge the process of the crisis by observing the fluctuation of asset prices.Due to the positive correlation between asset returns and risks,the rate of return on asset prices reflects the degree of risk to some extent.When the yield of an asset price accumulates to a certain extent,its high probability will have a price decline process.When asset prices rise significantly over time,we can assume that asset bubbles are forming and a crisis is likely to emerge.Changes in return on assets can be measured by the volatility of asset prices.Generally,the estimation and prediction model of the asset price can be divided into a parameter model and a non-parametric model according to whether the parameter is set,and can be divided into a low frequency model and a high frequency model according to the sampling frequency of the data.Common parametric models include the traditional GARCH model and the SV model,and this type of model is also commonly used in low frequency models.In the estimation of high-frequency data,the commonly used method is the realized volatility model of non-parametric estimation.Due to the high sampling frequency and non-parameter setting method,the high-frequency realized volatility model is in many aspects due to GARCH,SV and other models.Since the real world financial data may face the effects of microstructural noise and hopping in the market,the implemented covariance estimation method of highfrequency variance-covariance matrix is no longer effective,and the covariance matrix estimation of multiple assets may also face In order to make the estimation method of high frequency data volatility and covariance robust to jump and noise,many scholars have proposed many models based on the realized volatility.In this paper,the estimation of the volatility in the presence of noise is studied.The noise reduction models such as realized kernel,pre-averaging realized volatility and multi-time scale volatility are studied.The threshold volatility and the bio-power variation estimation are studied for the jump.After studying the theoretical background of the models,this paper conducts an empirical study of each estimation method,using the price of the 6-second tick asset of Ping An Bank(600000.SH)from 2009 to 2017.The noise reduction effect of each method is verified by comparing the mean values of the volatility obtained by various estimation methods,and an R/?V statistic is constructed to verify the rationality of each model setting.We find that the realized kernel,pre-average realized volatility,and multi-time scale volatility can reduce the impact of noise on volatility estimation to a certain extent.Among them,multi-time scale has achieved the best volatility noise reduction effect,the statistic R/?V shows that the volatility estimated by multiple time scales is also the most reasonable of the model settings.It is found that the realized threshold volatility and the bio-power variation can reduce the impact of the jump on the volatility estimation to a certain extent.For the estimation of the covariance matrix of multi-dimensional assets,we adopted the three methods of realized covariance,realized kernel covariance and realized threshold covariance for Ping An Bank(600000.SH)and Minsheng Bank(600016.SH).Empirical Research.The empirical results show that the kernel covariance has been implemented to reduce the influence of noise on the covariance matrix estimation.The threshold covariance method has been implemented to reduce the impact of the jump on the covariance matrix estimation process.Among the three,the best estimate is that the kernel covariance has been achieved.Finally,we use the method of kernel covariance and realized threshold covariance to estimate the covariance matrix of SSE 50 constituents,and use the obtained covariance matrix combined with risk parity to construct two highfrequency investments.Combining,by observing the performance of the high frequency combination and low frequency combination in 2009 to 2017,the portfolio based on the realized kernel covariance and the realized threshold covariance has obtained a lot of excess returns compared with the low frequency investment portfolio,indicating that High frequency data is more conducive to the management of portfolio than low frequency data.