We consider the least squares estimator of the classical AR(1) process when the underlying variables are the aggregated sums of independent random coefficient AR(1) models. We establish the asymptotics of the corresponding statistics and show that this estimator, in general, is not a consistent estimator of the expected value of autoregressive parameter when the sample size, or number of aggregating terms, or both tend to infinity. We propose a modified bias-corrected form leading to a consistent estimator.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados