Equity Portfolio Optimization using Latent Factor Models

Abstract

Portfolio managers use various portfolio strategies to maximize returns and minimize risks. We analyze the performances of a few of these strategies in the US and Chinese markets through backtesting. Our results find that strategies with covariance information perform better than those without, and that factor models perform the best out of the strategies tested, in both the US and Chinese markets. Using a Principal Component Analysis, we propose a new factor model by identifying latent factors that improve the Fama-French 3-factor model. Our results find that this new factor model performs better than both the Fama-French 3-factor and 5-factor models. Finally, we determine the number of latent factors for optimal portfolio performance. Our results find that 10 factors is optimal in the US market, and 20 factors is optimal in the Chinese market.