Forecasting realized volatility of BIST indices with har-type models

In this thesis, realized volatility of a selection of BIST Indices are forecasted with Heterogeneous Autoregressive Model (HAR) and its variations. For this purpose, ticks between 2001 and 2021 are used to generate 5-minute returns, which formed the basis for calculations of realized volatility and other realized measures. In the study, rolling windows are utilized for forecasting the volatility of one day ahead. These predictions are then compared to the actual realized volatilities. The thesis provides a thorough comparison of HAR-type models, and emphasizes the importance of underlying time series' characteristics in forecasting. Moreover, the findings of this thesis also hint at matters of diversification particular to index volatility forecasting. In overall, HAR Models proved to be a successful estimator for Turkish Stock Exchange time series.