1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
|
\name{Expectiles-Uniform}
\alias{Expectiles-Uniform}
\alias{eunif}
\alias{deunif}
\alias{peunif}
\alias{qeunif}
\alias{reunif}
\title{ Expectiles of the Uniform Distribution }
\description{
Density function, distribution function, and
expectile function and random generation for the distribution
associated with the expectiles of a uniform distribution.
}
\usage{
deunif(x, min = 0, max = 1, log = FALSE)
peunif(q, min = 0, max = 1, log = FALSE)
qeunif(p, min = 0, max = 1, Maxit.nr = 10, Tol.nr = 1.0e-6)
reunif(n, min = 0, max = 1)
}
%- maybe also 'usage' for other objects documented here.
\arguments{
\item{x, q}{
Vector of expectiles.
See the terminology note below.
}
\item{p}{
Vector of probabilities. % (tau or \eqn{\tau}).
These should lie in \eqn{(0,1)}.
}
\item{n, min, max, log}{
See \code{\link[stats:Uniform]{runif}}.
}
\item{Maxit.nr}{
Numeric.
Maximum number of Newton-Raphson iterations allowed.
A warning is issued if convergence is not obtained for all \code{p}
values.
}
\item{Tol.nr}{
Numeric.
Small positive value specifying the tolerance or precision to which
the expectiles are computed.
}
}
\details{
Jones (1994) elucidated on the property that the expectiles
of a random variable \eqn{X} with distribution function \eqn{F(x)}
correspond to the
quantiles of a distribution \eqn{G(x)} where
\eqn{G} is related by an explicit formula to \eqn{F}.
In particular, let \eqn{y} be the \eqn{p}-expectile of \eqn{F}.
Then \eqn{y} is the \eqn{p}-quantile of \eqn{G}
where
\deqn{p = G(y) = (P(y) - y F(y)) / (2[P(y) - y F(y)] + y - \mu),}{
p = G(y) = (P(y) - y F(y)) / (2[P(y) - y F(y)] + y - mu),}
and
\eqn{\mu}{mu} is the mean of \eqn{X}.
The derivative of \eqn{G} is
\deqn{g(y) = (\mu F(y) - P(y)) / (2[P(y) - y F(y)] + y - \mu)^2 .}{
g(y) = ( mu F(y) - P(y)) / (2[P(y) - y F(y)] + y - mu)^2 .}
Here, \eqn{P(y)} is the partial moment
\eqn{\int_{-\infty}^{y} x f(x) \, dx}{int^{y} x f(x) dx}
and
\eqn{0 < p < 1}.
The 0.5-expectile is the mean \eqn{\mu}{mu} and
the 0.5-quantile is the median.
A note about the terminology used here.
Recall in the \emph{S} language there are the \code{dpqr}-type functions
associated with a distribution, e.g.,
\code{\link[stats:Uniform]{dunif}},
\code{\link[stats:Uniform]{punif}},
\code{\link[stats:Uniform]{qunif}},
\code{\link[stats:Uniform]{runif}},
for the uniform distribution.
Here,
\code{unif} corresponds to \eqn{F} and
\code{eunif} corresponds to \eqn{G}.
The addition of ``\code{e}'' (for \emph{expectile}) is for the `other'
distribution associated with the parent distribution.
Thus
\code{deunif} is for \eqn{g},
\code{peunif} is for \eqn{G},
\code{qeunif} is for the inverse of \eqn{G},
\code{reunif} generates random variates from \eqn{g}.
For \code{qeunif} the Newton-Raphson algorithm is used to solve for
\eqn{y} satisfying \eqn{p = G(y)}.
Numerical problems may occur when values of \code{p} are
very close to 0 or 1.
}
\value{
\code{deunif(x)} gives the density function \eqn{g(x)}.
\code{peunif(q)} gives the distribution function \eqn{G(q)}.
\code{qeunif(p)} gives the expectile function:
the expectile \eqn{y} such that \eqn{G(y) = p}.
\code{reunif(n)} gives \eqn{n} random variates from \eqn{G}.
}
\references{
Jones, M. C. (1994)
Expectiles and M-quantiles are quantiles.
\emph{Statistics and Probability Letters},
\bold{20}, 149--153.
Yee, T. W. (2012)
Vector generalized linear and additive
quantile and expectile regression.
\emph{In preparation}.
}
\author{ T. W. Yee }
%\note{
%The ``\code{q}'', as the first character of ``\code{qeunif}'',
%may be changed to ``\code{e}'' in the future,
%the reason being to emphasize that the expectiles are returned.
%Ditto for the argument ``\code{q}'' in \code{peunif}.
%
%}
\seealso{
\code{\link{deexp}},
\code{\link{denorm}},
\code{\link{dunif}},
\code{\link{dkoenker}}.
}
\examples{
my_p <- 0.25; y <- runif(nn <- 1000)
(myexp <- qeunif(my_p))
sum(myexp - y[y <= myexp]) / sum(abs(myexp - y)) # Should be my_p
# Equivalently:
I1 <- mean(y <= myexp) * mean( myexp - y[y <= myexp])
I2 <- mean(y > myexp) * mean(-myexp + y[y > myexp])
I1 / (I1 + I2) # Should be my_p
# Or:
I1 <- sum( myexp - y[y <= myexp])
I2 <- sum(-myexp + y[y > myexp])
# Non-standard uniform
mymin <- 1; mymax <- 8
yy <- runif(nn, mymin, mymax)
(myexp <- qeunif(my_p, mymin, mymax))
sum(myexp - yy[yy <= myexp]) / sum(abs(myexp - yy)) # Should be my_p
peunif(mymin, mymin, mymax) # Should be 0
peunif(mymax, mymin, mymax) # Should be 1
peunif(mean(yy), mymin, mymax) # Should be 0.5
abs(qeunif(0.5, mymin, mymax) - mean(yy)) # Should be 0
abs(qeunif(0.5, mymin, mymax) - (mymin+mymax)/2) # Should be 0
abs(peunif(myexp, mymin, mymax) - my_p) # Should be 0
integrate(f = deunif, lower = mymin - 3, upper = mymax + 3,
min = mymin, max = mymax) # Should be 1
\dontrun{
par(mfrow = c(2,1))
yy <- seq(0.0, 1.0, len = nn)
plot(yy, deunif(yy), type = "l", col = "blue", ylim = c(0, 2),
xlab = "y", ylab = "g(y)", main = "g(y) for Uniform(0,1)")
lines(yy, dunif(yy), col = "darkgreen", lty = "dotted", lwd = 2) # 'original'
plot(yy, peunif(yy), type = "l", col = "blue", ylim = 0:1,
xlab = "y", ylab = "G(y)", main = "G(y) for Uniform(0,1)")
abline(a = 0.0, b = 1.0, col = "darkgreen", lty = "dotted", lwd = 2)
abline(v = 0.5, h = 0.5, col = "red", lty = "dashed") }
}
\keyword{distribution}
|
