Add invcrossmodelmatrix as field of VcovData

Project.toml

@@ -1,6 +1,6 @@
 name = "Vcov"
 uuid = "ec2bfdc2-55df-4fc9-b9ae-4958c2cf2486"
-version = "0.7.1"
+version = "0.8.0"
 
 [deps]
 Combinatorics = "861a8166-3701-5b0c-9a16-15d98fcdc6aa"

src/Vcov.jl

@@ -13,14 +13,16 @@ using Base: @propagate_inbounds
 ## Mimimum RegressionModel used in Vcov
 ##
 ##############################################################################
-struct VcovData{T, N} <: RegressionModel
+struct VcovData{T, Tu, N} <: RegressionModel
     modelmatrix::Matrix{Float64} # X
-    crossmodelmatrix::T          # X'X in the simplest case. Can be Matrix but preferably Factorization
+    crossmodelmatrix::T          # X'X 
+    invcrossmodelmatrix::Tu      # inv(X'X)
     residuals::Array{Float64, N} # vector or matrix of residuals (matrix in the case of IV, residuals of Xendo on (Z, Xexo))
     dof_residual::Int
 end
 StatsAPI.modelmatrix(x::VcovData) = x.modelmatrix
 StatsAPI.crossmodelmatrix(x::VcovData) = x.crossmodelmatrix
+invcrossmodelmatrix(x::VcovData) = x.invcrossmodelmatrix
 StatsAPI.residuals(x::VcovData) = x.residuals
 StatsAPI.dof_residual(x::VcovData) = x.dof_residual
 

src/covarianceestimators/vcovcluster.jl

@@ -93,6 +93,7 @@ function S_hat(x::RegressionModel, v::ClusterCovariance)
     # I choose the first option following reghdfe
     G = minimum(nclusters(v))
     rmul!(S, (size(modelmatrix(x), 1) - 1) / dof_residual(x) * G / (G - 1))
+    return Symmetric(S)
 end
 
 # res is a Vector in OLS, Matrix in IV
@@ -111,7 +112,7 @@ function helper_cluster(X::Matrix, res::Union{Vector, Matrix}, g::GroupedArray)
 end
 
 function StatsAPI.vcov(x::RegressionModel, v::ClusterCovariance)
-    xtx = inv(crossmodelmatrix(x))
+    xtx = invcrossmodelmatrix(x)
     pinvertible(Symmetric(xtx * S_hat(x, v) * xtx))
 end
 

src/covarianceestimators/vcovrobust.jl

@@ -28,7 +28,7 @@ function S_hat(x::RegressionModel, ::RobustCovariance)
 end
 
 function StatsAPI.vcov(x::RegressionModel, v::RobustCovariance)
-    xtx = inv(crossmodelmatrix(x))
+    xtx = invcrossmodelmatrix(x)
     pinvertible(Symmetric(xtx * S_hat(x, v) * xtx))
 end
 

src/covarianceestimators/vcovsimple.jl

@@ -11,12 +11,11 @@ Base.show(io::IO, ::SimpleCovariance) =
     print(io, "Simple covariance estimator")
 
 function S_hat(x::RegressionModel, ::SimpleCovariance)
-	rmul!(crossmodelmatrix(x), sum(abs2, residuals(x)))
+	Symmetric(crossmodelmatrix(x) .* sum(abs2, residuals(x)))
 end
 
 function StatsAPI.vcov(x::RegressionModel, ::SimpleCovariance)
-    invcrossmodelmatrix = Matrix(inv(crossmodelmatrix(x)))
-    rmul!(invcrossmodelmatrix, sum(abs2, residuals(x)) /  dof_residual(x))
-    Symmetric(invcrossmodelmatrix)
+    xtx = invcrossmodelmatrix(x)
+    Symmetric(xtx .* (sum(abs2, residuals(x)) /  dof_residual(x)))
 end
 

src/ranktest.jl

@@ -55,7 +55,7 @@ function ranktest!(X::Matrix{Float64},
         vlab = cholesky!(Hermitian((kronv * kronv') ./ size(X, 1)), check = false)
     else
         K = kron(Gmatrix, Fmatrix)'
-        vcovmodel = Vcov.VcovData(Z, K, X, size(Z, 1) - dof_small - dof_fes) 
+        vcovmodel = Vcov.VcovData(Z, K, X, nothing, size(Z, 1) - dof_small - dof_fes) 
         matrix_vcov2 = Vcov.S_hat(vcovmodel, vcov_method)
         vhat = K \ (K \ matrix_vcov2)'
         vlab = cholesky!(Hermitian(kronv * vhat * kronv'), check = false)