Forecasting stock returns with copula-GARCH model and an application on BIST30

In this thesis, copula-GARCH model, which is a combination of univariate Generalized Auto-Regressive Conditional Heteroskedasticity (GARCH) processes that take into account changing volatility / volatility clustering properties of financial asset returns and copula functions that are able to model multivariate nonlinear co-movement of returns, is employed to model stock returns listed in BIST30 Index and to obtain return forecasts by also taking into account multivariate dependence structure between the returns. Obtained returns from Elliptical and Mixed copula-based GARCH models are first employed in Global Minimum Variance, Tangency Portfolio and Global Minimum CVaR portfolio optimizations and performance of the constructed optimal portfolios are compared with the more traditional methods such as the equally weighted portfolio and an optimal portfolio constructed from historical data. Following the research on the application of returns obtained from copula-GARCH model in portfolio allocation problems, portfolio tail risk such as Value at Risk and Expected Shortfall forecasting performance of Elliptical and Regular Vine copula-GARCH models are evaluated by employing various backtesting procedures. According to the research results obtained first in optimal portfolio allocation and second in portfolio tail risk forecasting tasks, copula-GARCH models that take into account not only series specific properties but also various dependence types between the series performed well in both tasks. Especially, the ones employing copula functions that are able to model lower and upper tail dependencies are ranked at the top two best models which are either R-vine copula-GARCH and Student t copula-GARCH or mixed copula-GARCH and Student t copula-GARCH models.