New method accurately predicts future trends in complex time series data.
The article discusses how to estimate fractional time series models when the memory parameter is unknown and may be in different regions. The researchers show that the conditional sum of squares estimate is consistent and asymptotically efficient under Gaussianity. They also prove consistency and asymptotic normality for a general univariate model and asymptotic normality for a multivariate model using a one-step estimate.
