---------------------------------- Version 1.1 ----------------------------------

This version fixes a few bugs and adds parametric bootstrap functionality for
gpd modelling.

Changes from version 1.0:

1. bootgpd and methods added.
2. gpd was not passing 'start' through to gpd.fit. Fixed.
3. Rationalized summary.gpd and its methods into a single script (no impact).
4. summary.gpd did the wrong simulation when covariates were in the model. Fixed.
5. Bug in qqgpd caused x and y values to be reversed. The y-axis is scaled to 
   accomodate the simulated envelope and this meant that outliers could be
   excluded from the plot. Fixed.
6. print.summary.predict.mex was showing too many decimal places. Fixed
7. validate.texmex failed on Windows due to / in file path (all tests still
   passed if executed individually). Fixed.

---------------------------------- Version 1.2 ------------------------------------
Changes from version 1.1:

1. rl.gpd had a bug in the way it computed standard errors. Fixed.
2. Original Heffernan-Tawn methodology used Gumbel marginal distributions.
   Keef-Tawn-Papastathopoulos modified this to Laplace margins and identified 
   constraints that keep the model physically plausible. The modified and
   constrained approach is implemented and is now the default.
3. Diagnostic plot of constrained Profile Log-likelihood surface added, to 
   help users check for convergence to mle on constrained parameter space.
4. New diagnostic plots for GPD models when there are covariates, as 
   part of the plot method for a gpd object.
5. A returned gpd object now has residuals attached.

---------------------------------- Version 1.3 ------------------------------------

Changes from version 1.2:

1. predict method added for objects of class gpd, bgpd, bootgpd. This ought to
   make computation of return levels easier and more reliable. See the help
   files.
2. Bug fix. The 'Metropolis algorithm' as it was implemented was not actually
   a Metropolis algorithm due to a bug. The proposal distribution never changed
   from iteration to iteration. (Checks on the effects of the bug suggest it
   didn't produce distributions that were very different from those produced by
   an actual Metropolis algorithm.)
3. Cauchy distribution added as a proposal distribution. Defaults to Gaussian.
4. Speed improvement due to simulation of random numbers (and thus lots of
   matrix decomposition operations) being taken outside the simulation loop.
   The speed gain is largely offset by the need to run the chains for longer
   due to the bug fix described in item 2. Nevertheless, Dr. Metcalfe is a very
   clever boy.
5. Vignettes added. One for modelling clinical trial lab safety data; one for
   declustering.
6. NAMESPACE added.
7. Fixed bug in using linear predictor formulas when not passing a data.frame in.

---------------------------------- Version 2.1 ------------------------------------

Changes from version 1.2:

1.  Main modelling function switched from gpd to evm.
2.  Added Collate to DESCRIPTION.
3.  Covariance of predictions for GPD model does now NOT take account of the
    uncertainty in the rate, so estimated variances will be a little lower
    than in previous versions.
4.  GEV family added.
5.  Default fitting procedure is now MLE rather than weakly penalized MLE.
6.  Vignettes updated, switched from Sweave to knitr, multivariate vignette
    added.
7.  CITATION file added.
8.  All test functions and validate.texmex removed. These will reappear in a
    companion package.
9.  Nasty bug in predict functions fixed - xlevels added to output object from
    evm function.
10. Code refactoring.
11. NAMESPACE now keeps many internal functions internal.
12. An evm object now contains information in the true log-likelihood and the
    penalized log-likelihood. AIC uses the true log-likelihood by default.

---------------------------------- Version 2.3 -----------------------------------

Changes from version 2.1

1. egp3 family added together with vignette. Additional threshold selection plot
   based on egp3 added.
2. Test suite added back in, together with vignette.
3. Bug in u2gpd that was called from revTransform is now fixed. This affected
   plot.mex bootmex and a couple of other things.
4. Bootstrap of evm models is not parallelized.
5. AstraZeneca's support for the development of texmex is now credited in various
   places.
6. Available vignettes should now be listed from the package Index page.
7. Some instances of 'cat' and 'warning' have been changed to 'message' to allow
   for easier suppression and less annoyance.
8. Added Huber sandwich estimator of parameter estimate covariance matrix for 
   dependent data, for gpd model only at this stage.
9. Added functionality to carry out estimation of Joint Exceedance Curves, from 
   fitted Heffernan and Tawn model, documented in multivariate vignette.
10.Plotting now available using ggplot2 methods.

---------------------------------- Version 2.4 -----------------------------------

Changes from version 2.3

1. weibull family added.

---------------------------------- Version 2.4.2 -----------------------------------

Changes from version 2.4

1. gcpd family added (constrained parameter space for xi) as suggested by Yee &
   Stephenson (2007).
2. Profile likelihood based confidence interval estimation added for GPD models fit by
   evmOpt, only for models containing no covariates.
3. Added option for user to supply reference marginal distributions for marginal 
   transformation required in conditional dependence model estimation.  This is for use
   in instances where the marginal distributions are not well represented in the data 
   prividing joint observation (for instance where data represent only observaitons with
   a given variable above a threshold).
3. Bug in egp3RangeFit fixed (thanks to Emmanuel Flachire)
4. Bug fix: axis label in gpd range fit diagnostic plot.
5. Bug fix: plotting points in diagnostic plot for checkin independence of model residuals
   from conditioning varibale in mex fitting.
6. Various improvements to documentation.


