5/21/2009 version 1.8.1

The default calculation method for the Hausman test in relation
to the random effects estimator has been reversed: we now use 
the regression method by default, and the matrix-difference 
method only if the --matrix-diff option is given.  For the 
present, the --hausman-reg option has been retained but it
does not really do anything, it just confirms the default.

The command "lmtest" has been renamed as "modtest".  However,
"lmtest" has been retained as an alias so this is not a
true incompatibility.

The gretl reference manual says that the names of gretl
commands are reserved, and cannot be used as the names of 
variables.  This has not been properly enforced in the past,
but it is now.

The exponentation operator, '^', now associates rightward,
which is how it works in most mathematical software: e.g.
2^2^3 = 2^(2^3) = 256.

1/23/2009 version 1.8.0

The obsolete command "rhodiff" as been removed.  To create
a quasi-differenced series (as rhodiff did), use the "genr"
command.

The obsolete mechanism "save=" for saving residuals and/or
fitted values from system estimation has been removed; please
use the accessors $uhat and $yhat for this purpose.

The matrices returned by the accessors $sigma and $vcv for VAR
systems now have a degrees of freedom correction.

9/28/2008 version 1.7.9

The "values" function no longer automatically truncates its 
argument (series or vector) to integer.  To get the old 
behavior for an argument x you should do values(int(x)).

Generation of monthly dummy variables: these dummies are
now named "dm1", "dm2", ... , "dm12".  Previously we tried
to use 3-letter abbreviations for the names of the months in
the current locale, but this is too error-prone given
different possible character encodings and (new) support
for languages that do not use the Roman character set.

8/28/2008 version 1.7.7

Ordered probit and logit: the presentation of the cut-points in
ordered response models is now that used by Stata and R, not
by William Greene as previously.  That is, we don't estimate a
generic constant, but report n-1 cut points, where n is the
number of response levels.

7/30/2008 version 1.7.6

List arguments to user-defined functions:

In the context of a "foreach" loop across such a list, the
accessor $i still gets the name of the variable at position i 
in the list, but the variable is not "visible" by that name
within the function.  To get a handle on the variable (e.g.
for use on the right-hand side of a genr command), use the
new syntax listname.varname.  See the chapter on functions in
the User's Guide for details.

6/12/2008 version 1.7.5

$sigma and $vcv accessors:

The $sigma and $vcv accessors have been changed for some model 
types. Now $sigma always refers to a measure associated with the 
residual variance, while $vcv always refer to the variance of
the parameter estimates.

In the case of VARs/VECMs, $vcv formerly referred to the cross-
equation covariance of the residuals.  Now $sigma is used for
that purpose; $vcv gets the variance of the coefficients, which
was not previously accessible.

logical operators:

We are now enforcing what the gretl manual has said for quite
some time: logical OR is represented by "||" and logical AND
by "&&".  The single "|" is now reserved for row-wise matrix
concatenation, and the single "&" for "taking the address of"
a variable.  Previously we tried to guess when "|" and "&" were
intended to represent OR and AND respectively, as in the
original gretl syntax, but no more.

3/21/2008 version 1.7.4

The old "multiply" command has been removed.

The old commands "corc", "hilu" and "pwe" are consolidated
into a new one, "ar1".  The old command words remain as
aliases but their use is deprecated.

2/29/2008 version 1.7.3

You cannot define a user-function having the same name as
a built-in function.


