A1249
Title: High-order spectral estimation for mixed causal-non-causal and invertible-noninvertible (MARMA) models
Authors: Daniel Velasquez-Gaviria - Maastricht University (Netherlands)
Alain Hecq - Maastricht University (Netherlands) [presenting]
Abstract: The purpose is to explore the new methods for estimating parameters, identifying models, and simulating mixed causal-noncausal and invertible-noninvertible autoregressive moving average (MARMA) models driven by non-Gaussian noise. The proposed framework relies on high-order cumulants and combines the spectrum and the bispectrum into an estimation function. The global minimum of this estimation function accurately identifies the model that best fits the data while preserving the independent and identically distributed (iid) assumption for the estimated error sequences. To demonstrate the effectiveness of the proposed method, an extensive Monte Carlo study is conducted that shows unbiased estimated parameters and the ability to identify the correct model. Additionally, an empirical application is presented using the returns of 24 Fama-French stock portfolios of emerging markets. The results indicate that all the portfolios exhibit non-causal dynamics, resulting in strong white noise estimated residuals without conditional heteroscedastic effects.
