"Optimal Portfolio with Particle Filtering" (in Japanese)

This paper proposes a new method for constructing optimal portfolios with a particle filtering method, which shows we are able to improve performances of mean-variance portfolios substantially through estimation of expected returns and returns' volatilities based on Monte Carlo filter. In particular, we introduce state variables associated with expected returns as well as asymmetric volatilities in a state space framework and predict asset returns consistent with volatility changes in time. As a result, our estimated portfolios outperform not only mean-variance Portfolios with moving averages and variances of past returns, but also risk parity, minimum variance, and equally weighted portfolios, which do not depend on predictions of asset returns. Moreover, we construct portfolios with transaction costs and no-short-sale constraints, which possibly include Japanese REIT and U.S. REIT in addition to domestic and international bonds and equities with a riskless asset. Finally, performance evaluation based on accumulated returns, Sharpe ratios, Sortino ratios and maximum drawdowns confirms the validity of our method.