B1821
Title: Forecasting realized volatility of financial assets with limited historical data
Authors: Andreas Teller - Friedrich Schiller University Jena (Germany) [presenting]
Uta Pigorsch - University of Wuppertal (Germany)
Christian Pigorsch - Friedrich Schiller University Jena (Germany)
Abstract: The problem of forecasting realized volatility (RV) is considered for financial assets with limited historical data, such as new issues or spin-offs. Commonly, daily RV forecasting models rely on a sufficient history of data. For new issues and spin-offs, however, an extensive data history is not directly available. Therefore, it is proposed to forecast the RV of assets with limited historical data based on multi-source domain adaptation. Specifically, complementary source data of financial assets with a substantial historical data record is exploited by selecting source time series instances, that are most similar to the target data of the respective new issue or spin-off. Based on these instances and the target data, the heterogeneous autoregressive (HAR) model and modifications thereof are estimated, as well as feedforward neural network and extreme gradient boosting (XGBoost) models. Their forecasting performance is compared to forecasts of the same models but fitted exclusively to the target time series, as well as to a simplified pooling approach that includes the complete source and target data. The results indicate that the integration of complementary data can significantly improve the accuracy of RV forecasts, even shortly after their first trading day. In particular, the proposed instance selection regime shows superior performance compared to models based solely on target asset data or those that additionally incorporate the complete source data.
