MLE or LSE for AR(1) parameter. Sample mean correction used in MLE case. Intercept term estimated in LSE case.

Share:

Description

Fast exact computation of the MLE for AR(1) by solving the likelihood equation. The sample mean correction is used, so the method is not strickly speaking exact but the name derives from the fact that if the mean is known and was used instead of the sample mean the estimate would be an exact MLE estimate of the parameter in the AR(1) model. It has been shown that effect of estimating the sample mean is negligible.

Usage

1
ar1est(z, method = c("MLE", "LSE"))

Arguments

z

time series or vector

method

must be "MLE" or "LSE"

Details

The exact MLE for mean-zero an AR(1) time series satisfies a cubic equation. The solution of this equation for the MLE given by Zhang (2002) is used. This approach is more reliable as well as faster than the usual approach to the exact MLE using a numerical optimization technique which can occasionally have convergence problems.

Value

MLE for the parameter

Author(s)

A.I. McLeod and Ying Zhang

References

Zhang, Y. (2002). Topics in Autoregression, Ph.D. Thesis, University of Western Ontario.

Examples

1
2
3
4
5
6
7
8
9
#Example 1
#compare MLE and LSE for vel series
ar1est(vel)
ar1est(vel, method="MLE")
ar1est(vel, method="LSE")
#
#Example 2
ar1est(DiffBA)
ar1est(DiffBA, method="LSE")