Publication at Faculty of Social Sciences, Faculty of Mathematics and Physics, Centre for Economic Research and Graduate Education |
2021
Abstract
The directional news impact curve (DNIC) is a relationship between returns and the probability of next period's return exceeding a certain threshold-zero in particular. Using long series of S&P500 index returns and a number of parametric models suggested in the literature, as well and flexible semiparametric models, we investigate the shape of the DNIC and forecasting abilities of these models.
The semiparametric approach reveals that the DNIC has complicated shapes characterized by nonsymmetry with respect to past returns and their signs, heterogeneity across the thresholds, and changes over time. Simple parametric models often miss some important features of the DNIC, but some nevertheless exhibit superior out-of-sample performance.