forecastSNSTS-package: Forecasting of Stationary and Non-Stationary Time Series

In forecastSNSTS: Forecasting for Stationary and Non-Stationary Time Series

Description

Methods to compute linear h-step ahead prediction coefficients based on localised and iterated Yule-Walker estimates and empirical mean squared and absolute prediction errors for the resulting predictors. Also, functions to compute autocovariances for AR(p) processes, to simulate tvARMA(p,q) time series, and to verify an assumption from Kley et al. (2019).

Details

Package:	forecastSNSTS
Type:	Package
Version:	1.3-0
Date:	2019-09-02
License:	GPL (>= 2)

Contents

The core functionality of this R package is accessable via the function predCoef, which is used to compute the linear prediction coefficients, and the functions MSPE and MAPE, which are used to compute the empirical mean squared or absolute prediction errors. Further, the function f can be used to verify condition (10) of Theorem 3.1 in Kley et al. (2019) for any given tvAR(p) model. The function tvARMA can be used to simulate time-varying ARMA(p,q) time series. The function acfARp computes the autocovariances of a AR(p) process from the coefficients and innovations standard deviation.

Author(s)

Tobias Kley

References

Kley, T., Preuss, P. & Fryzlewicz, P. (2019). Predictive, finite-sample model choice for time series under stationarity and non-stationarity. Electronic Journal of Statistics, forthcoming. [cf. https://arxiv.org/abs/1611.04460]

forecastSNSTS documentation built on Sept. 2, 2019, 5:06 p.m.