Mean-Variance Portfolio with Generalized EMA and Anomaly Detector

Abstract

Listen

Translate article

This paper proposes a novel scheme for achieving high investment performances with Mean-Variance (MV) portfolios. As is well-known, MV portfolio performances largely depend on the quality of estimates of parameters, namely expected returns and covariance matrices. Particularly, easily implementable exponential moving average (EMA) for expected returns with rolling variances (RV) and correlations are frequently used in practice.To obtain better estimates leading to higher performances, we utilize a particle filtering method for a newly developed generalized exponential moving average and stochastic volatility model (GEMASV model). Moreover, in order for excluding model mis-specifications, we effectively apply the filtering-based anomaly detector to demonstrate resulting performances substantially improve the ones with the standard EMA and RV.