release 0.0.5

Author: zhouxiaoyu

README.md

@@ -1,8 +1,6 @@
 FixedEffectModel: A Python Package for Linear Model with High Dimensional Fixed Effects.
 =======================
-<img width="150" src="https://user-images.githubusercontent.com/61887305/126601384-35ab02e9-447c-4977-8d29-aa89397727bd.png" alt="Logo" />
-
-
+<img width="223" alt="image" src="https://user-images.githubusercontent.com/61887305/154887555-abd074bc-e69d-4793-b850-94a0ce437b01.png">
 
 [![Downloads](https://pepy.tech/badge/fixedeffectmodel)](https://pepy.tech/project/fixedeffectmodel)
 [![Downloads](https://pepy.tech/badge/fixedeffectmodel/month)](https://pepy.tech/project/fixedeffectmodel)
@@ -271,6 +269,29 @@ Return an object of results:
 
 Return the same object of results as fixedeffect does.
 
+We also provide two-step GMM estimator if you set thet option "gmm2=True".
+Define a matrix <img src="https://latex.codecogs.com/svg.image?P_Z&space;=&space;Z(Z'Z)^{-1}Z'" title="P_Z = Z(Z'Z)^{-1}Z'" />
+
+- "ivgmm", the one-step GMM estimator generate 
+<img src="https://latex.codecogs.com/svg.image?\begin{equation}\beta_{2SLS}&space;=&space;(X'&space;P_Z&space;X)^{-1}(X'&space;P_Z&space;y)\end{equation}" title="\begin{equation}\beta_{2SLS} = (X' P_Z X)^{-1}(X' P_Z y)\end{equation}" /> with variance-covariance matrices equal
+  - Unadjusted. Define <img src="https://latex.codecogs.com/svg.image?\sigma_u^2=u'u/df" title="\sigma_u^2=u'u/df" />, the variance-covariance matrix is <img src="https://latex.codecogs.com/svg.image?Var(\beta)=\sigma_u^2[X'Z(Z'Z)^{-1}Z'X]^{-1}" title="Var(\beta)=\sigma_u^2[X'Z(Z'Z)^{-1}Z'X]^{-1}" />
+  - Heteroskedasticity robust. Define <img src="https://latex.codecogs.com/svg.image?\hat{\Omega}=&space;diag(\hat{u}_1^2,\dotsm,&space;\hat{u}_N^2)" title="\hat{\Omega}= diag(\hat{u}_1^2,\dotsm, \hat{u}_N^2)" /> and <img src="https://latex.codecogs.com/svg.image?W_2&space;=&space;Z'\hat{\Omega}&space;Z" title="W_2 = Z'\hat{\Omega} Z" />
+  , the variance-covariance matrix is <img src="https://latex.codecogs.com/svg.image?Var_r(\beta)=[X'P_z&space;X]^{-1}[X'Z(Z'Z)^{-1}&space;W_2&space;(Z'Z)^{-1}&space;Z'X][X'&space;P_z&space;X]^{-1}" title="Var_r(\beta)=[X'P_z X]^{-1}[X'Z(Z'Z)^{-1} W_2 (Z'Z)^{-1} Z'X][X' P_z X]^{-1}" />
+  - Cluster. Deine <img src="https://latex.codecogs.com/svg.image?W_2&space;=&space;\sum_g&space;Z_g'u_g&space;u_g'&space;Z_g" title="W_2 = \sum_g Z_g'u_g u_g' Z_g" />
+  , the variance-covariance matrix is  <img src="https://latex.codecogs.com/svg.image?Var_c(\beta)=[X'P_z&space;X]^{-1}[X'Z(Z'Z)^{-1}&space;W_2&space;(Z'Z)^{-1}&space;Z'X][X'&space;P_z&space;X]^{-1}" title="Var_c(\beta)=[X'P_z X]^{-1}[X'Z(Z'Z)^{-1} W_2 (Z'Z)^{-1} Z'X][X' P_z X]^{-1}" />
+- "ivgmm" with "gmm2=True", the two-step GMM estimator generate <img src="https://latex.codecogs.com/svg.image?\begin{equation}\beta_{GMM}&space;=&space;[X'&space;Z&space;W_2^{-1}Z'&space;X]^{-1}[X'&space;Z&space;W_2^{-1}Z'&space;Xy]\end{equation}" title="\begin{equation}\beta_{GMM} = [X' Z W_2^{-1}Z' X]^{-1}[X' Z W_2^{-1}Z' Xy]\end{equation}" />
+  - Unadjusted. <img src="https://latex.codecogs.com/svg.image?Var(\beta_{GMM})&space;=&space;(X'ZW_2^{-1}Z'X)^{-1}" title="Var(\beta_{GMM}) = (X'ZW_2^{-1}Z'X)^{-1}" />
+  - Heteroskedasticity robust. Define <img src="https://latex.codecogs.com/svg.image?W_3=Z'\Omega_2&space;Z" title="W_3=Z'\Omega_2 Z" />
+  and <img src="https://latex.codecogs.com/svg.image?\Omega_2" title="\Omega_2" /> as the 
+  diagonal matrix generated using the residual from the two-step GMM.
+  , the variance-covariance matrix is <img src="https://latex.codecogs.com/svg.image?\begin{equation}Var_r(\beta_{GMM})=[X'Z(W_2)^{-1}Z'X]^{-1}&space;[X'Z(W_2)^{-1}W_3(W_2)^{-1}Z'X]&space;[X'Z(W_2)^{-1}Z'X]^{-1}\end{equation}" title="\begin{equation}Var_r(\beta_{GMM})=[X'Z(W_2)^{-1}Z'X]^{-1} [X'Z(W_2)^{-1}W_3(W_2)^{-1}Z'X] [X'Z(W_2)^{-1}Z'X]^{-1}\end{equation}" />
+  - Cluster. Define 
+
+  <img src="https://latex.codecogs.com/svg.image?\begin{aligned}&W_2&space;=&space;\sum_g&space;Z_g'u_g&space;u_g'&space;Z_g\\&W_3&space;=&space;\sum_g&space;Z_g'u_{2,g}&space;u_{2,g}'&space;Z_g\\\end{aligned}" title="\begin{aligned}&W_2 = \sum_g Z_g'u_g u_g' Z_g\\&W_3 = \sum_g Z_g'u_{2,g} u_{2,g}' Z_g\\\end{aligned}" />
+
+  , the variance-covariance matrix is <img src="https://latex.codecogs.com/svg.image?\begin{equation}Var_c(\beta_{GMM})=[X'Z(W_2)^{-1}Z'X]^{-1}&space;[X'Z(W_2)^{-1}W_3(W_2)^{-1}Z'X]&space;[X'Z(W_2)^{-1}Z'X]^{-1}\end{equation}" title="\begin{equation}Var_c(\beta_{GMM})=[X'Z(W_2)^{-1}Z'X]^{-1} [X'Z(W_2)^{-1}W_3(W_2)^{-1}Z'X] [X'Z(W_2)^{-1}Z'X]^{-1}\end{equation}" />.
+
+
 ### DID
 *model = did (data_df = None, dependent = None, exog_x = None, treatment = None, csid = None, tsid = None, exp_date = None, group_effect = 'treatment', cluster = None, formula = None, robust = False, noint = False, c_method = 'cgm', psdef = True)*
 
@@ -370,6 +391,7 @@ result.summary()
 ivtest(result)
 
 #------------------------------#
+
 #define fixed effect model
 exog_x = ['x_1']
 y = ['y']

fixedeffect/_version.py

@@ -1 +1 @@
-__version__ = '0.0.4'
+__version__ = '0.0.5'

fixedeffect/iv/iv.py

@@ -13,18 +13,19 @@
 import pandas as pd
 import warnings
 
+
 class iv2sls:
     def __init__(self,
                  data_df,
-                 dependent = None,
-                 exog_x = None,
-                 endog_x = [],
-                 iv = [],
+                 dependent=None,
+                 exog_x=None,
+                 endog_x=[],
+                 iv=[],
                  category=[],
                  cluster=[],
-                 formula = None,
-                 robust = False,
-                 noint = False,
+                 formula=None,
+                 robust=False,
+                 noint=False,
                  **kwargs
                  ):
         """
@@ -42,6 +43,7 @@ def __init__(self,
         **kwargs:some hidden option not supposed to be used by user
         """
 
+        warnings.warn('"iv2sls" will no longer be supported beyond version 0.0.5, please use "ivgmm" instead')
 
         # grammar check
         if (exog_x is None) & (formula is None):
@@ -59,8 +61,8 @@ def __init__(self,
                 raise NameError('You have to input endogenous variables for iv2sls')
 
         else:
-            dependent, exog_x, category_input, cluster_input, endog_x, iv = dependent, exog_x, category,  \
-                                                                              cluster, endog_x, iv
+            dependent, exog_x, category_input, cluster_input, endog_x, iv = dependent, exog_x, category, \
+                                                                            cluster, endog_x, iv
 
         # df preprocess
         data_df.fillna(0, inplace=True)
@@ -80,8 +82,8 @@ def __init__(self,
         self.orignal_exog_x = orignal_exog_x
 
     def fit(self,
-            epsilon = 1e-8,
-            max_iter = 1e6):
+            epsilon=1e-8,
+            max_iter=1e6):
 
         data_df = self.data_df
         dependent = self.dependent
@@ -115,7 +117,7 @@ def fit(self,
             for i in iv:
                 all_cols.append(i)
             all_cols.append(dependent[0])
-            demeaned_df = demean_dataframe(data_df, all_cols, category_input, epsilon = epsilon, max_iter = max_iter)
+            demeaned_df = demean_dataframe(data_df, all_cols, category_input, epsilon=epsilon, max_iter=max_iter)
             if noint is False:
                 for i in all_cols:
                     demeaned_df[i] = demeaned_df[i].add(data_df[i].mean())
@@ -127,9 +129,8 @@ def fit(self,
         iv_model = []
         iv_result = []
         endog_x_hat = []
-        
 
-        #-----------------  First stage -------------------#
+        # -----------------  First stage -------------------#
         # if OLS on raw data:
         if noint is False:
             const_x_first_stage = sm.add_constant(demeaned_df[x_first_stage], has_constant='add')
@@ -155,10 +156,10 @@ def fit(self,
         x_second_stage = exog_x + endog_x_hat
         x_second_stage_original = exog_x + endog_x
 
-        #----------------- Second stage  -----------------#
+        # ----------------- Second stage  -----------------#
         # if OLS on raw data:
         if noint is False:
-            #x_second_stage_ = sm.add_constant(demeaned_df[x_second_stage])
+            # x_second_stage_ = sm.add_constant(demeaned_df[x_second_stage])
             x_first_stage = ['const'] + x_first_stage
             x_second_stage = ['const'] + x_second_stage
             x_second_stage_original = ['const'] + x_second_stage_original
@@ -173,7 +174,7 @@ def fit(self,
         n = demeaned_df.shape[0]
         k = len(x_second_stage)
 
-        #------ initiate result object ------#
+        # ------ initiate result object ------#
         f_result = OLSFixed()
         f_result.model = 'iv2sls'
         f_result.dependent = dependent
@@ -197,14 +198,12 @@ def fit(self,
         else:
             std_error = result.bse * np.sqrt((n - k) / (n - k + k0 - rank))  # for fe if k0=1 need to add it back
         f_result.bse = std_error
-        
 
         # compute summary statistics and save in result
         self.compute_summary_statistics(result, f_result, rank)
 
         return f_result
 
-
     def compute_summary_statistics(self,
                                    result,
                                    f_result,
@@ -216,7 +215,6 @@ def compute_summary_statistics(self,
 
         x_second_stage = f_result.x_second_stage
 
-
         if self.noint is True:
             k0 = 0
         else:
@@ -247,7 +245,6 @@ def compute_summary_statistics(self,
         else:
             raise NameError('Total sum of square equal 0, program quit.')
 
-
         f_result.rsquared_adj = 1 - (len(data_df) - k0) / (result.df_resid) * (1 - f_result.rsquared)
 
         if k0 == 0:
@@ -282,7 +279,6 @@ def compute_summary_statistics(self,
         else:
             f_result.fvalue = 0
 
-
         f_result.f_pvalue = f.sf(f_result.fvalue, result.df_model, f_result.df)
         f_result.f_df_proj = [result.df_model, f_result.df]