add method for matrices

[PaulSoderlind]

like vcov(::HAC,Jacobian::Matrix,momentCondition::Matrix;... )

Motivation: this would allow users to bypass DataFrames and GLM, but still use CovarianceMatrices.jl, which could increase the user base of the CovarianceMatrices.jl package.

Example based on the first linear regression in README.md:

bw = 2

v = vcov(BartlettKernel(bw),lm1,prewhite = false)

X     = [ones(T) x]
(T,K) = (size(X,1),size(X,2))
b     = X\y
res   = y - X*b

B   = inv(X'X)                                    #Jacobian of moment conditions
A   = lrvar(BartlettKernel(bw),res.*X)*T*(1+K/T)   #df adjustment is built into vcov
vPS = B*A*B

v and vPS are virtually the same.

As an alternative to create a method, this could perhaps just be added to the documentation.

[gragusa]

I have been thinking about this for a while. The alternative that I am exploring is to have a package called MomentConditionProblem where one defines MomentCondition and JacobianMomentCondition and then provide an interface to CovarianceMatrices.

[gragusa]

@PaulSoderlind, I think I have finished working on the master branch and I have done everything I wanted to do. The only thing left is your request. How would you like something like this:

sandwich(::HAC, H::Matrix, momentmatrix::Matrix; dof::Float64 = 0.0)

where dof is 0.0 (no adjustment) or k, in which case k degrees of freedom are subtracted to n.

vcov is something that I don't like (at least should be acov for asymptotic covariance). sandwich is a possibility.

Note: This is now implemented.