Dynamic Asset Allocation Strategy Study Based On Bayesian Method

Asset allocation,especially dynamic asset allocation,is an important factor in determining returns for an investment portfolio.However,dynamic asset allocation problem is not easy to solve.Firstly,the assets have some correlation and the variance of assets changes rapidly over time,so estimating the time-varying covariances is a difficult task.Secondly,risky asset return generation process is of significance in optimal asset allocation,but we do not know the real data generation process.Furthermore,investors do not have complete information of all relevant parameters of the asset return model.If ignoring parameter uncertainty and model uncertainty,asset allocation decision will be sun-optimal.To solve the difficulties in dynamic asset allocation research,this paper builds a dynamic asset allocation analysis model based on Bayesian method.On the one hand,aiming at the problem of time-varying covariance matrix parameter estimation,this paper applies principal component analysis to construct orthogonal eigen-portfolios,which reduces the data dimension and reduces the number of estimated parameters.On the other hand,aiming at asset return generation process uncertainty and incomplete information and the asymmetric effects of asset returns,this paper takes the threshold effect,leverage effect and Dirichlet process mixed model into the stochastic volatility model,namely the Dirichlet process mixed threshold stochastic volatility model(TSVDPM)and the Dirichlet process mixed threshold stochastic volatility model with leverage effect(TSVL-DPM).We use the Markov Chain Monte Carlo(MCMC)and particle learning(PL)to estimate the parameters of the model and solve the asset allocation problem.In this paper,we apply the TSV-DPM model and the TSVL-DPM model to the daily returns of Gem comprehensive index and Shanghai Stock Exchange composite index of China.Empirical results show that there exists a high persistence of volatility and a significant leverage effect and strong evidence of asymmetries in Chinese stock market.Specifically,the volatility persistence tends to be higher,and both volatility of volatility tends to be lower when bad news impacts the market than when good news of the same magnitude does.In addition,it is found that the MCMC and PL methods have similar estimation accuracy,and the PL method is obviously superior to the MCMC method in calculating the cost.Finally,we take three types of assets,namely China's four Exchange Traded Funds as samples,to conduct empirical research on the non-parametric Bayesian dynamic asset allocation model.Using the log predictive tail score(LPTS)to compare the predictive power of the asset return generation model in extreme events,the empirical study suggests that the TSVL-DPM has the best predictive power.Using the Sharpe ratio to evaluate the asset allocation model,the empirical study suggests that the nonparametric Bayesian dynamic asset allocation strategy is obviously better than the parameter asset allocation strategy.