Title: Portfolio optimization based on forecasts from vine copula GARCH models using external regressors
Authors: Maziar Sahamkhadam - Linnaeus University (Sweden)
Andreas Stephan - Jonkoping University (Sweden)
Ralf Ostermark - Abo Akademi (Finland)
Andreas Stephan - Linnaeus University (Sweden) [presenting]
Abstract: Based on AR-GARCH-Vine copula models, we forecast and simulate one-day-ahead returns of twelve industry portfolios. We add Fama-French factors to the mean and the volatility index (VIX) to the volatility equation as external regressors to the AR-GARCH model. For modeling the dependency structure between assets, we use regular, drawable and canonical vine copulas. Copula families include Clayton, Student-t and a mix of Gaussian, Student-t, Clayton and Gumbel with all rotations. We also consider skewed t and generalized extreme value distributions for modeling univariate marginals. We extend previous studies and provide estimates using the selection and estimation algorithm for regular vine copula. In general, the portfolio strategies based on maximizing the Sharpe ratio (CET) are outperforming the ones with minimizing conditional value-at-risk (MinCVaR). For CET portfolio strategies, skewed t distribution (as the marginal distribution) shows higher out-of-sample economic performance in comparison to generalized extreme value distribution (GEV). In terms of accumulated returns, portfolio strategies based on regular vine copulas are outperforming the strategies based on canonical and drawable vine copulas. In general, for MinCVaR portfolios, there is not much difference when changing vine copula family.