A1589
Title: Regressions with heavy tailed weakly nonstationary processes
Authors: Ioannis Kasparis - University of Cyprus (Cyprus) [presenting]
Abstract: The interaction of long memory/persistence with heavy tails results in an enlargement of the nonstationary region, i.e. the covariate model space for which conventional inference is not applicable. Parametric and non-parametric regression methods are considered to bridge inference between stationary and nonstationary environments in the presence of heavy tails. A new limit theory is first developed for heavy-tailed weakly nonstationary processes (HT-WNPs hereafter), i.e. processes that lie on the threshold of nonstationarity. It is then shown that the proposed methods yield conventional inference for a wide range of heavy-tailed covariates, including stationary long memory, WNPs, and strongly nonstationary long memory. Possible applications to the predictability of stock returns by risk measures are provided.