def create_table(models, out):
"""Creates a table of regression results.
Args:
models: regression results
out: the generated table is saved here
Returns:
None
"""
results = []
model_names = []
covariate_names = {}
for model in models:
with open(model, 'rb') as file:
result = (pickle.load(file))
results.append(result)
model_names.append(result.model_name)
covariate_names.update(result.var_names)
table = Stargazer(results)
table.dependent_variable_name(
covariate_names[results[0].model.endog_names])
table.custom_columns(model_names, [1] * len(model_names))
table.rename_covariates(covariate_names)
latex_table = table.render_latex()
latex_table = re.sub(r"l(c+)\}", r"lc\1}", latex_table)
# ugly hack because stargazer generates an invalid latex table
with open(out, 'w') as file:
file.write(latex_table)
def Appendix_Table_3(df):
df_short_g = pd.DataFrame({
"beliefadjustment_normalized":
df[df["dummynews_goodbad"] == 0][df[df["dummynews_goodbad"] ==
0]['treatgroup'] ==
4]['beliefadjustment_normalized'],
"beliefadjustment_bayes_norm":
df[df["dummynews_goodbad"] == 0][df[df["dummynews_goodbad"] == 0]
['treatgroup'] == 4]
['beliefadjustment_bayes_norm']
})
model_ols = smf.ols(
formula="beliefadjustment_normalized ~ beliefadjustment_bayes_norm",
data=df_short_g)
reg_s_1 = model_ols.fit(cov_type='HC1')
df_short_b = pd.DataFrame({
"beliefadjustment_normalized":
df[df["dummynews_goodbad"] == 1][df[df["dummynews_goodbad"] ==
1]['treatgroup'] ==
4]['beliefadjustment_normalized'],
"beliefadjustment_bayes_norm":
df[df["dummynews_goodbad"] == 1][df[df["dummynews_goodbad"] == 1]
['treatgroup'] == 4]
['beliefadjustment_bayes_norm']
})
model_ols = smf.ols(
formula="beliefadjustment_normalized ~ beliefadjustment_bayes_norm",
data=df_short_b)
reg_s_2 = model_ols.fit(cov_type='HC1')
df["interact_negative_bayes"] = df["beliefadjustment_bayes_norm"] * df[
"dummynews_goodbad"]
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ beliefadjustment_bayes_norm + dummynews_goodbad + interact_negative_bayes",
data=df[df['treatgroup'] == 4])
reg_s_3 = model_ols.fit(cov_type='HC1')
Appendix_Table_3 = Stargazer([reg_s_1, reg_s_2, reg_s_3])
Appendix_Table_3.title('Table 8: Belief Adjustment in the Short-Run')
Appendix_Table_3.dependent_variable_name('Belief Adjustment')
Appendix_Table_3.custom_columns([
'Positive Information', 'Negative Information',
'Difference-in-difference'
], [1, 1, 1])
return Appendix_Table_3
def Main_Table_1(df):
df_good = pd.DataFrame({
"beliefadjustment_normalized":
df[df["dummynews_goodbad"] == 0]['beliefadjustment_normalized'],
"dummytreat_direct1month":
df[df["dummynews_goodbad"] == 0]['dummytreat_direct1month'],
"rank":
df[df["dummynews_goodbad"] == 0]['rank'],
"beliefadjustment_bayes_norm":
df[df["dummynews_goodbad"] == 0]['beliefadjustment_bayes_norm']
})
model_ols = smf.ols(
formula="beliefadjustment_normalized ~ dummytreat_direct1month",
data=df_good)
reg_1 = model_ols.fit(cov_type='HC1')
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + rank + beliefadjustment_bayes_norm",
data=df_good)
reg_2 = model_ols.fit(cov_type='HC1')
df_bad = pd.DataFrame({
"beliefadjustment_normalized":
df[df["dummynews_goodbad"] == 1]['beliefadjustment_normalized'],
"dummytreat_direct1month":
df[df["dummynews_goodbad"] == 1]['dummytreat_direct1month'],
"rank":
df[df["dummynews_goodbad"] == 1]['rank'],
"beliefadjustment_bayes_norm":
df[df["dummynews_goodbad"] == 1]['beliefadjustment_bayes_norm']
})
model_ols = smf.ols(
formula="beliefadjustment_normalized ~ dummytreat_direct1month",
data=df_bad)
reg_3 = model_ols.fit(cov_type='HC1')
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + rank + beliefadjustment_bayes_norm",
data=df_bad)
reg_4 = model_ols.fit(cov_type='HC1')
#Generating interaction term
df["interact_direct1month"] = df["dummytreat_direct1month"] * df[
"dummynews_goodbad"]
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + dummynews_goodbad + interact_direct1month",
data=df)
reg_5 = model_ols.fit(cov_type='HC1')
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + dummynews_goodbad + rank + interact_direct1month + beliefadjustment_bayes_norm",
data=df)
reg_6 = model_ols.fit(cov_type='HC1')
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + dummynews_goodbad + interact_direct1month + rankdummy1 + rankdummy2 + rankdummy3 + rankdummy4 + rankdummy5 + rankdummy6 + rankdummy7 + rankdummy8 + rankdummy9 + rankdummy1_interact + rankdummy2_interact + rankdummy3_interact + rankdummy4_interact + rankdummy5_interact + rankdummy6_interact + rankdummy7_interact + rankdummy8_interact + rankdummy9_interact",
data=df)
reg_7 = model_ols.fit(cov_type='HC1')
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + dummynews_goodbad + interact_direct1month + beliefadjustment_bayes_norm + rankdummy1 + rankdummy2 + rankdummy3 + rankdummy4 + rankdummy5 + rankdummy6 + rankdummy7 + rankdummy8 + rankdummy9 + rankdummy1_interact + rankdummy2_interact + rankdummy3_interact + rankdummy4_interact + rankdummy5_interact + rankdummy6_interact + rankdummy7_interact + rankdummy8_interact + rankdummy9_interact",
data=df)
reg_8 = model_ols.fit(cov_type='HC1')
Main_Table_1 = Stargazer(
[reg_1, reg_2, reg_3, reg_4, reg_5, reg_6, reg_7, reg_8])
Main_Table_1.title(
'Table 1 - Belief Adjustment: Direct versus One Month Later')
Main_Table_1.dependent_variable_name('Normalized Belief Adjustment - ')
Main_Table_1.custom_columns([
'Positive Information', 'Negative Information',
'Difference-in-difference',
'Difference-in-difference with rank fixed effects'
], [2, 2, 2, 2])
return Main_Table_1
def Appendix_Table_1(df):
df_good = pd.DataFrame({
"beliefadjustment_normalized":
df[df["dummynews_goodbad_h"] == 0]['beliefadjustment_normalized'],
"dummytreat_direct1month":
df[df["dummynews_goodbad_h"] == 0]['dummytreat_direct1month'],
"rank":
df[df["dummynews_goodbad_h"] == 0]['rank'],
"beliefadjustment_bayes_norm":
df[df["dummynews_goodbad_h"] == 0]['beliefadjustment_bayes_norm']
})
model_ols = smf.ols(
formula="beliefadjustment_normalized ~ dummytreat_direct1month",
data=df_good)
reg_1 = model_ols.fit(cov_type='HC1')
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + rank + beliefadjustment_bayes_norm",
data=df_good)
reg_2 = model_ols.fit(cov_type='HC1')
df_bad = pd.DataFrame({
"beliefadjustment_normalized":
df[df["dummynews_goodbad_h"] == 1]['beliefadjustment_normalized'],
"dummytreat_direct1month":
df[df["dummynews_goodbad_h"] == 1]['dummytreat_direct1month'],
"rank":
df[df["dummynews_goodbad_h"] == 1]['rank'],
"beliefadjustment_bayes_norm":
df[df["dummynews_goodbad_h"] == 1]['beliefadjustment_bayes_norm']
})
model_ols = smf.ols(
formula="beliefadjustment_normalized ~ dummytreat_direct1month",
data=df_bad)
reg_3 = model_ols.fit(cov_type='HC1')
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + rank + beliefadjustment_bayes_norm",
data=df_bad)
reg_4 = model_ols.fit(cov_type='HC1')
#Generating interaction term
df["interact_direct1month"] = df["dummytreat_direct1month"] * df[
"dummynews_goodbad"]
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + dummynews_goodbad_h + interact_direct1month",
data=df)
reg_5 = model_ols.fit(cov_type='HC1')
model_ols = smf.ols(
formula=
"beliefadjustment_normalized ~ dummytreat_direct1month + dummynews_goodbad_h + rank + interact_direct1month + beliefadjustment_bayes_norm",
data=df)
reg_6 = model_ols.fit(cov_type='HC1')
Appendix_Table_1 = Stargazer([reg_1, reg_2, reg_3, reg_4, reg_5, reg_6])
Appendix_Table_1.title(
'Appendix Table 1 - Belief Adjustment: Direct versus One Month Later')
Appendix_Table_1.dependent_variable_name('Normalized Belief Adjustment - ')
Appendix_Table_1.custom_columns([
'Positive Information', 'Negative Information',
'Difference-in-difference'
], [2, 2, 2])
return Appendix_Table_1
def ols_regression_formatted(data,
specifications,
as_latex=False,
covariates_names=None,
covariates_order=None):
"""
Creates formatted tables for different dependent variables and specifications
Input:
data (df): Dataframe containing all necessary variables for OLS regression
specifications (dictionary): dependent variables as keys and list of specifications
as values
as_latex (bool): specify whether Output as table or Latex code
covariate_names (dict): dictionary with covariate names as in "data" as keys and new
covariate names as values
Output:
list_of_tables (list of stargazer tables): list of formatted tables
"""
# Create dictionary which connects dependent variables with formatted tables
dict_regression_tables = {}
# Generate regressions
for depvar in specifications.keys():
regression_list = []
specification_list = specifications[depvar]
list_all_covariates = []
for specification in specification_list:
estimation_equation = depvar + " ~ " + specification
regression = smf.ols(data=data, formula=estimation_equation).fit()
regression_list.append(regression)
# Create set of all variables for this dependent variable
list_all_covariates = list(
set(list_all_covariates +
regression.params.index.values.tolist()))
# Format table with stargazer
formatted_table = Stargazer(regression_list)
# No dimension of freedoms and blank dependent variable
formatted_table.show_degrees_of_freedom(False)
formatted_table.dependent_variable_name("")
# Optional: Change order of covariates
if covariates_order is not None:
covariates_order_depvar = list(
OrderedSet(covariates_order).intersection(list_all_covariates))
list_remaining_covariates = list(
OrderedSet(list_all_covariates).difference(
OrderedSet(covariates_order_depvar)))
covariates_sorted = list(
OrderedSet(covariates_order_depvar).union(
list_remaining_covariates))
covariates_sorted.remove("Intercept")
covariates_sorted = covariates_sorted + ["Intercept"]
formatted_table.covariate_order(covariates_sorted)
# Optional: Change name of covariates
if covariates_names is not None:
formatted_table.rename_covariates(covariates_names)
# Add table or latex code to dictionary
if as_latex is True:
dict_regression_tables[depvar] = formatted_table.render_latex()
# Delete tabular environment around it
dict_regression_tables[depvar] = dict_regression_tables[
depvar].replace("\\begin{table}[!htbp] \\centering\n", "")
dict_regression_tables[depvar] = dict_regression_tables[
depvar].replace("\\end{table}", "")
else:
dict_regression_tables[depvar] = formatted_table
return dict_regression_tables
