Theoretical And Empirical Analysis Of Hedging And Portfolio Optimization Based On Different Frequency Data

With the rapid progress of modern computer science and information technology and the deepening of economic globalization, financial markets have been increasingly developing in recent decades. The Internet and other communication technology-based electronic trading markets have become the dominant organizational form of financial markets. The costs of data storage and data processing continue to decrease. Mass financial market data is properly collected, stored and processed. The frequency, accuracy and reliability of the data are also increasing. Traditional market statistics is mainly recoded in day, week, month, year frequency, called the low-frequency data. High frequency data is mainly referred to intraday data collected in second, minute, or hour frequency. Compared to low frequency data, high frequency data is a finer record of the transaction process of securities, which contains important details ignored by low-frequency data. Thus, high frequency data is an edged tool to uncover short-term price dynamics. High frequency data plays a more and more important role in understanding the features of the specific securities or asset classes. It also helps investors discover innovative investment opportunities and trading strategies, deduce transaction costs and increase trading flexibility, improve risk management and decision-making and so forth. High frequency data is widely used in many developed financial markets, where various intraday trading strategies based on high-frequency data have been invented.Currently high-frequency data plays a role in the futures markets in China, while the main stock exchanges do not allow the intraday turnover trading. Nevertheless, high-frequency data contains a lot of information about securities price volatility, which can be used to improve various investment decisions. In this context, this dissertation will combine the low frequency data and high frequency data to analyze the price processes of representative stock indexes and stock index futures, to investigate the interaction between asset return and volatility. This dissertation also studies other important financial decision-making problems such as the dynamic hedging strategy for stock portfolio, the construction and evaluation of dynamic portfolio policy. The main contributions of this dissertation include:(1) Under the general Ito semimartingale framework, nonparametric model specification tests are implemented to study features of asset price processes. To ensure the validity of the tests, both the market microstructure effects and the possibility of observational equivalence should be treated carefully. The following measures have been taken:selecting representative securities with strong liquidity for empirical study, using a large sample of ultrahigh-frequency data, and checking the stability of empirical results with respect to free parameters of the tests. Empirical results show that the price process of CSI 300 Stock Index Futures contains diffusion process, finite activity jump process and infinite activity jump process as its components.(2) In order to examine the complex interactions between asset return and volatility, a stochastic volatility model with stochastic leverage (SV-SL model) is proposed. The SV-SL model is a nonlinear state space model, which extends the standard stochastic volatility model with constant leverage (SV-L model). In order to estimate the SV-SL model efficiently, an efficient importance sampling-based simulated maximum likelihood estimation method is proposed in which the daily realized volatility is used as a proxy for the latent volatility. The proposed model and estimation approach are applied to two Chinese stock market indexes, namely the Shanghai Stock Exchange composite index and Shenzhen Stock Exchange component index. The empirical results suggest no evidence of significant leverage effect in Chinese stock markets. However, significant stochastic leverage effect in Chinese stock markets are discovered. Furthermore, the direction of leverage effect is closely related to the performance of stock market. Specifically, leverage effect exists in "bear" markets, while reverse leverage effect exists in "bull" markets. These findings reconcile conflicts between empirical results based on standard stochastic volatility model and those based on GARCH Models.(3) The single-period stock portfolio hedging strategies and relevant performance evaluation methods are thoroughly studies from the perspective of decision theory. The implement of minimum variance hedge strategies depends on the econometric models of spot prices and futures prices. Three sets of modeling variables, i.e., price changes, simple returns and logarithmic returns, can be considered equivalent in theory, but the minimum variance hedge ratios should be computed according to different formulae. With the help of high frequency data, new estimation methods of optimal hedge ratio and new measures of hedging performance are proposed. Based on statistical decision theory and multiple model evaluation methods, a new inference framework for comparing multiple hedging strategies is proposed.(4) Taking the daily mark-to-market system into consideration, a multiperiod hedge model with futures is constructed under the mean-variance framework. Time inconsistency under the multiple mean-variance frameworks is treated by considering two kinds of optimal hedging policies, i.e., time-consistent optimal hedging strategy and pre-commitment optimal hedging strategy. The recursive formulae of both optimal strategies are established. With the help of further restrictions on the dynamic process of futures price and spot price, the optimal hedging strategy is greatly simplified so that the policy is independent of the utility function under the classic expected utility maximum framwork. Meanwhile, the importance of overall planning during the hedging period is illustrated by the comparative analysis of three different policies, i.e., the dynamic optimal hedging strategy, the myopic optimal hedging strategy, and the static optimal hedging strategy.(5) Considering possible intertemporal statistical correlations among asset returns, multiperiod portfolio optimization models are constructed under the mean-variance framework. Both the recursive formulae of the time-consistent optimal portfolio strategy with riskless asset and without riskless asset are established. Assuming linear predictability of asset returns, empirical studies are designed to comparing two different multiperiod portfolio strategies, i.e., the optimal strategy considering the intertemporal statistical correlations and the optimal strategy assuming no intertemporal statistical correlations. Under the representative specifications of parameters including initial wealth, investment horizon, risk aversion, and riskless interest rate, etc., the empirical distribution of the terminal wealth for a multiperiod portfolio strategy is obtained by numeric simulation methods, and then two portfolio performance measures, i.e., the certainty equivalence of the terminal wealth and the Sharp ratio, are computed. Empirical results show that intertemporal statistical correlations among asset returns cannot be ignored and the optimal strategy considering the statistical correlations can bring significant economic value.