Abstract:

This study aims to determine the most effective model for forecasting volatility within the Nordic stock markets. In this regard, the forecasting power of HAR-RV, RSV, and PS models is compared to the ARFIMA-RV model using high frequency data for 7 Nordic stock market indices spanning from 2010 to 2019. One-day-ahead out-of-sample realized volatility forecasts are produced using a recursive window mechanism. The out-of-sample forecast losses are measured by the MSE and QLIKE criteria. The results indicate several noteworthy points. Firstly, the HAR-RV (PS and RSV) models are suggested to be best performing realized volatility models over the ARFIMA-RV model. Secondly, the separation of realized variance into positive and negative realized semivariances, which is known as good and bad volatilities, might offer valuable financial insights in certain situations, aiding the prediction of future realized volatility. Lastly, the results and findings are specific to market, data frequency, time horizon, and some characteristics of data, emphasizing the importance of these factors in interpreting the findings.

Keywords: