Monthly downloads



Gamma and Exponential Generalized Linear Models with Elastic Net Penalty

Implements the fast iterative shrinkage-thresholding algorithm (FISTA) algorithm to fit a Gamma distribution with an elastic net penalty as described in Chen, Arakvin and Martin (2018) <arxiv:1804.07780>. An implementation for the case of the exponential distribution is also available, with details available in Chen and Martin (2018) <https://papers.ssrn.com/abstract_id=3085672>.


Build Status CRAN\_Status\_Badge Downloads


This package provides an implementation of the elastic net penalty for Gamma and exponentially distributed response variables.


You can install the stable version on R CRAN.

install.packages("RPEGLMEN", dependencies = TRUE)

You can install the development version from GitHub.


Background Information

This package is designed to provide the user to fit an Exponential or Gamma distribution to the response variable with an elastic net penalty on the predictors. This package is of particular use in combination with the RPEIF and RPESE packages, in which the influence function of a time series of returns is used to compute the standard error of a risk and performance measure. See Chen and Martin (2018) for more details.

For the computational details to fit a Gamma distribution on data with an elastic net penalty, see Chen, Arakvin and Martin (2018).


# Sample Code

# Load the package

# Function to return the periodogram of data series
myperiodogram <- function (data, max.freq = 0.5, twosided = FALSE, keep = 1){
  data.fft <- fft(data)
  N <- length(data)
  tmp <- Mod(data.fft[2:floor(N/2)])^2/N
  tmp <- sapply(tmp, function(x) max(1e-05, x))
  freq <- ((1:(floor(N/2) - 1))/N)
  tmp <- tmp[1:floor(length(tmp) * keep)]
  freq <- freq[1:floor(length(freq) * keep)]
  if (twosided) {
    tmp <- c(rev(tmp), tmp)
    freq <- c(-rev(freq), freq)
  return(list(spec <- tmp, freq <- freq))

# Function to compute the standard error based the periodogram of the influence functions time series
SE.Gamma <- function(data, d = 7, alpha = 0.5, keep = 1, exponential.dist = TRUE){
  # Compute the periodograms
  my.periodogram <- myperiodogram(data)
  my.freq <- my.periodogram$freq
  my.periodogram <- my.periodogram$spec
  # Remove values of frequency 0 as it does not contain information about the variance
  my.freq <- my.freq[-1]
  my.periodogram <- my.periodogram[-1]
  # Implement cut-off
  nfreq <- length(my.freq)
  my.freq <- my.freq[1:floor(nfreq*keep)]
  my.periodogram <- my.periodogram[1:floor(nfreq*keep)]
  # GLM with BFGS optimization
  # Create 1, x, x^2, ..., x^d
  x.mat <- rep(1,length(my.freq))
  for(col.iter in 1:d){
    x.mat <- cbind(x.mat,my.freq^col.iter)
  # Fit the Exponential or Gamma model
    res <- glmnet_exp(x.mat, my.periodogram, alpha.EN = alpha) else
      res <- fit.glmGammaNet(x.mat, my.periodogram, alpha.EN = alpha)
  # Return the estimated variance

# Loading hedge fund data from PA
data(edhec, package <- "PerformanceAnalytics")

# Computing the expected shortfall for the time series of returns
test.mat <- apply(edhec, 2, IF.ES)
test.mat <- apply(test.mat, 2, as.numeric)

# Returning the standard errors from the Exponential distribution fit
apply(test.mat, 2, SE.Gamma, exponential.dist = TRUE)


This package is free and open source software, licensed under GPL (>= 2).

Functions in RPEGLMEN

Name Description
glmnet_exp Elastic Net Penalized Exponentially Distributed Response Variables
fit.glmGammaNet Elastic Net Penalized Gamma or Exponentially Distributed Response Variables
No Results!

Vignettes of RPEGLMEN

No Results!

Last month downloads


Type Package
Date 2019-08-30
License GPL (>= 2)
Encoding UTF-8
LazyData true
LinkingTo Rcpp, RcppEigen
RoxygenNote 6.1.1
NeedsCompilation yes
Biarch true
SystemRequirements C++11
VignetteBuilder R.rsp
Packaged 2019-09-05 20:45:21 UTC; antho
Repository CRAN
Date/Publication 2019-09-07 10:10:02 UTC

Include our badge in your README