Title: Portfolio optimization based on dynamic factor and dynamic conditional correlation GARCH models
Authors: Maziar Sahamkhadam - Linnaeus University (Sweden) [presenting]
Andreas Stephan - Linnaeus University (Sweden)
Ralf Ostermark - Abo Akademi (Finland)
Abstract: Dynamic Factor Analysis (DFA) is used in the context of portfolio optimization based on Dynamic Conditional Correlation (DCC)-GARCH forecasting models. We use Gaussian State-Space modeling and Kalman filtering to extract the hidden state variables (factors), assuming the standard identifiable model and parameter constraints. The hidden factors are then inserted into the mean equation of DCC-GARCH and ARMA-DCC-GARCH models to perform out-of-sample forecasts. Having obtained mean and volatility forecasts, we simulate one-day ahead returns from multivariate normal distribution and allocate optimal weights based on three utility functions including maximum Sharpe ratio (CET), Global Minimum Variance (GMV) and minimum Conditional Value-at-Risk (Min-CVaR). We apply the DFA-DCC-GARCH and DFA-ARMA-DCC-GARCH models to portfolios which consist of twelve U.S. industry indexes or ten stock indexes and compare them with simple DCC-GARCH models as the benchmarks. In general, for both stock indexes and industry indexes, the models based on DFA outperform the benchmarks either in terms of maximizing the investor's utility function (maximizing Sharpe ratio or minimizing CVaR) or increasing the accumulation wealth. However, there is not much improvement for GMV portfolios. In conclusion, the DFA-based models are suitable for both CET and Min-CVaR portfolios.