v__0.190

mec/mx.py

@@ -3,6 +3,27 @@
 import numpy as np, scipy.sparse as sp, pandas as pd
 
 
+def ivgmm(Y_i,X_i_k,Z_i_l,W_l_l=None):
+    I = len(Y_i)
+    if W_l_l is None:
+        W_l_l = np.linalg.inv( Z_i_l.T @ Z_i_l /  I )
+    Pi_i_i =  Z_i_l @ W_l_l @ Z_i_l.T
+    beta_k = np.linalg.solve(X_i_k.T @ Pi_i_i @ X_i_k,  X_i_k.T @ Pi_i_i @ Y_i)
+    val = (Y_i - X_i_k @ beta_k).T @ Pi_i_i @ (Y_i - X_i_k @ beta_k) / (2* I * I)
+    return beta_k,val
+
+def efficient_ivgmm(Y_i,X_i_k,Z_i_l, centering = True):
+    I=len(Y_i)
+    beta1_k = ivgmm(Y_i,X_i_k,Z_i_l )[0] # first stage obtained by 2SLS
+    epsilon1_i = Y_i - X_i_k @ beta1_k
+    mhat_l_i = Z_i_l.T * epsilon1_i[None,:]
+    mbar_l = mhat_l_i.mean(axis=1)
+    Sigmahat1_l_l = (mhat_l_i @ mhat_l_i.T) / I - centering * mbar_l[:,None] * mbar_l[None,:]
+    W_l_l =  np.linalg.inv(Sigmahat1_l_l)
+    beta2_k, val2 = ivgmm(Y_i,X_i_k,Z_i_l,W_l_l )
+    return beta2_k,val2
+
+
 def iv_gmm(Y_i,X_i_k,Z_i_l, efficient=False, centering = True):
     def beta_gmm(Y_i,X_i_k,Z_i_l,W_l_l=None ):
         ZtildeT_k_i =  (X_i_k.T @ Z_i_l  @ Z_i_l.T) if W_l_l is None else (X_i_k.T @ Z_i_l @ W_l_l @ Z_i_l.T)
@@ -21,8 +42,3 @@ def beta_gmm(Y_i,X_i_k,Z_i_l,W_l_l=None ):
     XPiY_k_i = X_i_k.T @ Pi_i_i @ Y_i
     objval = (Y_i.T @ Pi_i_i @ Y_i - XPiY_k_i.T @ np.linalg.inv(  X_i_k.T @ Pi_i_i @ X_i_k ) @ XPiY_k_i )/ (2*I*I)
     return beta_k,objval
-
-
-
-
-

setup.py

@@ -2,7 +2,7 @@
 
 setup(
 	name="mec",
-	version="0.189",
+	version="0.190",
 	authors=["Alfred Galichon"],
 	author_email="ag133@nyu.edu",
 	licence="",