Over-Accepted Causality Hypothesis: Misspecification of Models, Missing Filters or Mimic Processes?

Misspecification of Models, Missing Filters or Mimic Processes?

Abstract

In order to verify the strictly non linear nature of causal relationship, a
procedure suggested by Bekiros and Diks (2008) was utilized, resorting
to consecutive filtering of residuals via various processes. After
application of a GARCH-BEEK filter, remaining significant
unidirectional nonlinear causal relationship was found running from
exchange rate to oil prices. Thus, we investigate whether rejection of null
hypothesis is due to misspecification of used GARCH model or other
reasons. Hence, a Mackey Glass filter was used to eliminate the bias
caused by deterministic chaos which resulted in the acceptance of null
hypothesis according to which oil prices do not cause variations in
exchange rate and vice versa. It can be concluded that, firstly, its
plausible that Mackey Glass function, in filtering processes, takes
slightly into account fluctuations that can be classified as conditional
heteroskedasticity. Secondly, chaotic structure amplified by a white noise
process might mimic a conditionally Heteroskedastic one, which can
explain the reaction of the nonlinear causality test after GARCH
filtering