hw06 upload

assignments/did/code/did-133463.py

@@ -0,0 +1,109 @@
+# Librerias
+import os
+import numpy as np
+import pandas as pd
+from linearmodels import PanelOLS
+
+# Carga de datos
+PATH = os.path.join('..', 'data', 'callaway-santanna.csv')
+df = pd.read_csv(PATH)
+
+# Renombrar columnas
+df.rename(columns={'year':'t', 'countyreal':'i', 'first.treat':'treat_start'}, inplace=True)
+
+# NA a "treat_start" si i en control
+county_treatment_detail = (
+    df
+    .groupby(['i'], as_index=False)['treat']
+    .apply(lambda x: 1 if x.sum() > 0 else 0)
+)
+df['treat_start'] = (
+    np
+    .select(
+        [
+            (df['i'].isin(county_treatment_detail[county_treatment_detail['treat']==0]['i']))
+        ],
+        [
+            np.nan
+        ],
+        default=df['treat_start']
+    )
+)
+df['k'] = df['t'] - df['treat_start']
+
+# Multi-index
+df.set_index(['i', 't'], inplace=True)
+
+# Dummies
+df = (
+    pd
+    .get_dummies(
+        data=df, 
+        columns=['k'], 
+        prefix='k', 
+        prefix_sep='_',
+        dummy_na=True,
+        dtype='int'
+    )
+)
+
+# Modelo 0
+treated_pop = (df['treat']==1)
+m0 = (
+    PanelOLS(
+        dependent=df[treated_pop]['lemp'],
+        exog=df[treated_pop][[x for x in df.columns if 'k' in x and 'k_nan' not in x and 'k_-1.0' not in x]],
+        entity_effects=True,
+        time_effects=True,
+        drop_absorbed=True
+    )
+)
+res0 = m0.fit(cov_type='clustered')
+
+# Test de Wald - Modelo 0
+R0 = (
+    np
+    .array(
+        [
+            [1, 0, 0, 0, 0, 0],
+            [0, 1, 0, 0, 0, 0],
+            [0, 0, 1, 0, 0, 0]
+        ]
+    )
+)
+V = (
+    np
+    .array(
+        [0, 0, 0]
+    )
+)
+f0 = res0.wald_test(R0, V)
+anticipation0 = False
+att0 = '-'
+
+# Modelo 1
+m1 = (
+    PanelOLS(
+        dependent=df['lemp'],
+        exog=df[[x for x in df.columns if 'k' in x and 'k_nan' not in x and 'k_-1.0' not in x]],
+        entity_effects=True,
+        time_effects=True,
+        drop_absorbed=True
+    )
+)
+res1 = m1.fit(cov_type='clustered')
+
+# Test de Wald - Modelo 1
+R1 = (
+    np
+    .array(
+        [
+            [1, 0, 0, 0, 0, 0, 0],
+            [0, 1, 0, 0, 0, 0, 0],
+            [0, 0, 1, 0, 0, 0, 0]
+        ]
+    )
+)
+f1 = res1.wald_test(R1, V)
+anticipation1 = False
+att1 = '-'