engleski

Prediction accuracy of market risk forecasts: GARCH-type models towards volatility benchmark

The Volatility, considered as a quantified measure of market risk, plays an important role in numerous financial applications. Estimating and forecasting volatility is extremely important not only for investors but also for risk managers. Our research focuses on the problem of volatility forecasting in the financial markets. The main objective is to determine a prediction accuracy of the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) family models towards volatility benchmark as well as to examine in which periods GARCH-type models underestimate or overestimate the true volatility and why. We compare the forecast accuracy of several GARCH-type models commonly used in financial risk management. As a true variance of returns so called integrated variance (IV) is unknown, IV is consistently estimated by realized variance (RV). Unlike the common practice widely employed in empirical research where RV is obtained using daily returns over quite a long period of time, we calculated a daily realized volatility using optimal sampling returns and employ the daily measures to forecast future volatility. When optimal sampling frequency is found, RV is calculated and declared as a benchmark for a comparison purpose. Finding a benchmark based on high frequency data enable us to determine prediction accuracy of volatility models proposed in the literature. The aim of our research is to select optimal sampling frequency for calculating realized variance and to examine if models using high frequency data obtain better returns forecasts compared to models using day-to-day returns. The forecasting accuracy of the models is tested using the Croatian main stock market index CROBEX for the period from January 4, 2010 to April 28, 2017.

market risk, Volatility forecasting, High-frequency data, GARCH models, CROBEX

nije evidentirano