loexp: Expectile regression for time series curve

In expectile: Modelling of expectiles

Description

Expectile regression for time series curve by decomposing data matrix Y = X B + E, where B is "mostly non-negative".

Usage

loexp(y, w=rep(1, length(y)), sigma=40, polyo=2, alpha=0.5, biweight=4.685, tol=1e-04, maxIter=50)

Arguments

y	A numeric time series vector of length T.
w	A numeric vector of T case weights. If unspecificed, all weights are set to 1.
sigma	The standard deviation of Gaussian kernel.
polyo	The order of polynomial, currently can be 0, 1, or 2.
alpha	The desired expectile.
biweight	Parameter used in Tukey's biweight function.
tol	The tolerance for expectile estimation.
maxIter	The maximum number of iterations in estimation step.

Value

A list with components:

intparams	An integer vector parameters: return status (0=,1=,2=,3=), length of input vector, polyo and maxit.
dblparams	A double vector of parameters: sigma, alpha, biweight and tolerance.
y	The input vector of data.
w	The input vector of case weights.
outy	The output vector of expectiles for the input data y.
outw	The output vector of weights used in fitting.

Author(s)

Asa Wirapati, Mark Robinson

References

[1] ...

Examples

data(loexp.example)

plot(ex1, pch=19, type="b")

# 50th expectile (median)
lines(loexp(ex1, alpha=0.5)$outy, lwd=4, col="blue")

# 0.5th expectile (baseline)
lines(loexp(ex1, alpha=0.005)$outy, lwd=4, col="red")

plot(ex2.y, pch=19, type="b")
# give weight=0 for 0s
lines(loexp(ex2.y, w=ex2.w, alpha=0.005)$outy, lwd=4, col="blue")

expectile documentation built on May 2, 2019, 6:11 p.m.