A1304
Title: Bayesian nonparametric portfolio selection with rolling maximum drawdown control
Authors: Mei Xiaoling - Xiamen University (China) [presenting]
Abstract: The portfolio selection problem is considered for a multiperiod investor who seeks to maximize mean-variance utility facing multiple risky assets and various trading constraints in the presence of return predictability. With the presence of trading constraints, dynamic programming is impractical to carry out due to the curse of dimensionality. Model predictive control is implemented, which is computationally efficient to solve the problem, and it is proposed to use a nonparametric Bayesian model, i.e., hierarchical Dirichlet process-based Hidden Markov Model (HDP-HMM) to predict the multiperiod mean and covariance of returns. Then, a time-varying maximum drawdown is considered to adjust the risk aversion, which can efficiently cope with the limit loss problems. Both simulation and empirical results show that trading strategies based on the proposed method can provide better out-of-sample performance than the existing methods.
