MODELLING AND FORECASTING VOLATILITY OF SECTOR INDICES ON ZAGREB STOCK EXCHANGE: MULTIVARIATE GARCH MODEL

The goal of this paper is modelling and forecasting volatility of sector indices on Zagreb Stock Exchange i.e. on the Croatian capital market using a multivariate generalized autoregressive conditional heteroscedastic (MGARCH) model according to the two assumptions about conditional distributions of returns, i.e. Gaussian (normal) and Student’s t-distribution and to determine whether these distributions are appropriate when describing the distribution of returns of sector indices on the Zagreb Stock Exchange. When making investment decisions, investors take into account the expected return and investment risk, and are assisted by analyses related to various risk measures such as volatility of returns, correlation coefficients between returns, etc. Daily data on returns of sector indices on Zagreb Stock Exchange are analysed within time span from 2013 to 2021 and encompass the sectors of industrial production, construction, production and processing of food and tourism. For the sake of the analysis, MGARCH is used to estimate dynamic conditional correlation (DCC-GARCH) models. Calculated unconditional volatility indicates greater stability of the tourism sector compared to other sectors, while unconditional correlation coefficients indicate a slight strength of the relationship between the sectors and thus the possibility of achieving diversification effects when investing. Conditional volatilities of sector indices returns are moving close, except the construction sector, whose volatility increases over time indicating a growing presence of risk in the sector. Furthermore, existance of a raising trend in conditional correlation of returns especially when the sector of tourism is included, suggests the growing connection of other sec- tors with this particular one. That connection is especially visible between sectors of tourism and production and processing of food. Model comparisons indicate that MGARCH model based on Student t-distribution is more appropriate for describing the returns’ distribution of sector indices in relation to the model based on Gaussian distribution. The analysis provided in the paper has some limitations that might be observed in relatively short time span of model assessment and specifics of Croatian capital market compared with developed markets.