def placebo_reg(df_placebo):
# store (exogenous) regressors for first and second stage in a list
regr = ['const', 'mean_hvac_share', 'mean_unempl', 'munempldiff', 'mskattekrdiff', 'mbefdiff', 'ekbistandpc_diff', 'dsmastad',
'dstorstad', 'dsv_maj_1', 'dseatsmp_1', 'dseatsop_1', 'flyktingandel_tplus1', 'flyktingandel_tplus2', 'flyktingandel_tplus3']
exog = ['const', 'mean_hvac_share', 'mean_unempl', 'munempldiff', 'mskattekrdiff', 'mbefdiff', 'ekbistandpc_diff', 'dsmastad',
'dstorstad', 'dsv_maj_1', 'dseatsmp_1', 'dseatsop_1']
# first stage
reg_placebo1 = mt.reg(df_placebo, y_name = 'nonOECDshare_diff_tplus1', x_name = regr, cluster = 'kommun')
reg_placebo2 = mt.reg(df_placebo, y_name = 'nonOECDshare_diff_tplus2', x_name = regr, cluster = 'kommun')
reg_placebo3 = mt.reg(df_placebo, y_name = 'nonOECDshare_diff_tplus3', x_name = regr, cluster = 'kommun')
# second stage
iv_placebo = mt.ivreg(df_placebo, y_name = 'soc_bidr_diff', x_name = ['nonOECDshare_diff_tplus1', 'nonOECDshare_diff_tplus2',
'nonOECDshare_diff_tplus3'], z_name = ['flyktingandel_tplus1', 'flyktingandel_tplus2', 'flyktingandel_tplus3'], w_name = exog,
iv_method = '2sls', cluster = 'kommun')
# call function from auxiliary file that creates LaTex table fragments for Table 6
get_table6(df_placebo, reg_placebo1, reg_placebo2, reg_placebo3, iv_placebo)
# perform joint test of the three placebo treatments in the second stage regression
# return p-value of the test
pvalue = iv_placebo.Ftest(['nonOECDshare_diff_tplus1', 'nonOECDshare_diff_tplus2', 'nonOECDshare_diff_tplus3'])[1]
return(reg_placebo1, reg_placebo2, reg_placebo3, pvalue)
def main_reg(df):
# store (exogenous) regressors for first and second stage in a list
regressors_base = ['const', 'mean_hvac_share', 'mean_unempl', 'munempldiff', 'mskattekrdiff', 'mbefdiff', 'ekbistandpc_diff',
'dsmastad', 'dstorstad', 'dsv_maj_1', 'dseatsmp_1', 'dseatsop_1', 'panel8891', 'panel9194', 'flyktingandel']
exog_base = ['const', 'mean_hvac_share', 'mean_unempl', 'munempldiff', 'mskattekrdiff', 'mbefdiff', 'ekbistandpc_diff', 'dsmastad',
'dstorstad', 'dsv_maj_1', 'dseatsmp_1', 'dseatsop_1', 'panel8891', 'panel9194']
### standard errors clustered at municipality level
# first stage
reg1 = mt.reg(df, y_name = 'nonOECDshare_diff', x_name = regressors_base, cluster = 'kommun')
# second stage
iv1 = mt.ivreg(df, y_name = 'soc_bidr_diff', x_name = 'nonOECDshare_diff', z_name = 'flyktingandel', w_name = exog_base,
iv_method = '2sls', cluster = 'kommun')
### standard errors clustered at county level
# first stage
reg2 = mt.reg(df, y_name = 'nonOECDshare_diff', x_name = regressors_base, cluster = 'countykod')
# second stage
iv2 = mt.ivreg(df, y_name = 'soc_bidr_diff', x_name = 'nonOECDshare_diff', z_name = 'flyktingandel',
w_name = exog_base, iv_method = '2sls', cluster = 'countykod')
# call function from auxiliary file that creates Latex table fragments for Table 2
get_table2(df, reg1, iv1, reg2, iv2)
# perform F-test for relevance of the instrument
F_stat1 = reg1.Ftest('flyktingandel')[0]
F_stat2 = reg2.Ftest('flyktingandel')[0]
return(reg1, iv1, reg2, iv2, F_stat1, F_stat2)
def reg1(year,data,ss,control):
for i in data["year"]:
if i == year:
reg= mt.reg(data.loc[(data["year"]==i)], # DataFrame to use
'lis_share', # Outcome
pd.concat([data[[ss]],control]), # Indep. Variables
fe_name='wkr', # Fixed-effects
cluster='wkr') # Cluster
return reg.beta[ss], reg.se[ss], reg.r2, reg.N,reg.Ftest(ss)
def setup_class(cls):
"""Stata reg output from `sysuse auto; reg price mpg`"""
test_path = path.split(path.relpath(__file__))[0]
auto_path = path.join(test_path, 'data', 'auto.dta')
autodata = pd.read_stata(auto_path)
y = 'price'
x = ['mpg', 'length']
cls.result = reg(autodata, y, x, cluster='gear_ratio', addcons=True)
cls.expected = ols_cluster
def reg4(year,data,ss,control):
for i in data["year"]:
if i == year:
reg= mt.reg(data.loc[(data["year"]==i)], # DataFrame to use
"dNazi_share", # Outcome
pd.concat([data[[ss]],control]), # Indep. Variables
fe_name="wkr", # Fixed-effects
cluster='wkr')
return reg.beta[ss],reg.se[ss], reg.N,reg.r2
def setup_class(cls):
"""Stata reg output from `sysuse auto; reg price mpg`"""
cls.init(cls)
cls.precision['vce'] = 6
test_path = path.split(path.relpath(__file__))[0]
auto_path = path.join(test_path, 'data', 'auto.dta')
autodata = pd.read_stata(auto_path)
y = ['price']
x = ['mpg', 'length']
cls.result = reg(autodata, y, x, addcons=True)
cls.expected = ols_std
def ghat_of_x(y, x, x0, h, degree, kernel):
K = kernel_func(x - x0, h, kernel)
if _sparse_data(K, degree):
return np.nan
X = _make_X(x, x0, degree)
x_name = ['cons'] + ['x{}'.format(i) for i in range(1, degree + 1)]
df = pd.DataFrame(np.hstack((y.reshape(-1, 1), X, K.reshape(-1, 1))),
columns=['y'] + x_name + ['k'])
res = reg(df, 'y', x_name, awt_name='k')
beta = res.beta
# plot_this(y, x, K, X, res) # XXX tmp, diagnostic
return beta['cons']
def reg2(year,data,ss,control):
for i in data["year"]:
if i == year:
locat = ["pops", "pops_2", "pops_3", "pops_4", "pops_5","bigd","city","altitude"]
socioe= ["c25juden_share","c25kath_share", "c25arbei_share", "c25anges_share", "war_per1000","in_welfare_per1000", "sozialrentner_per1000" ,"logtaxprop"]
elect = ["turnout24" ,"vote24", "nsfb24" , "spd24","z24"]
reg= mt.reg(data.loc[(data["year"]==i)], # DataFrame to use
ss, # Outcome
control, # Indep. Variables
fe_name="wkr", # Fixed-effects
cluster='wkr') # Cluster
return reg.beta["bigd"], reg.se["bigd"],reg.beta["altitude"], reg.se["altitude"],\
reg.Ftest(locat), reg.Ftest(socioe),reg.Ftest(elect),\
reg.r2, reg.N
def sens1_reg(df, df_sens1):
# run regression without dummy for large-sized municipal population as we excluded big city counties
# otherwise all values of that variable are zero and we cannot invert the regressor matrix
regressors_sens1 = ['const', 'mean_hvac_share', 'mean_unempl', 'munempldiff', 'mskattekrdiff', 'mbefdiff', 'ekbistandpc_diff', 'dsmastad',
'dsv_maj_1', 'dseatsmp_1', 'dseatsop_1', 'panel8891', 'panel9194', 'flyktingandel']
exog_sens1 = ['const', 'mean_hvac_share', 'mean_unempl', 'munempldiff', 'mskattekrdiff', 'mbefdiff', 'ekbistandpc_diff', 'dsmastad',
'dsv_maj_1', 'dseatsmp_1', 'dseatsop_1', 'panel8891', 'panel9194']
# first stage
reg_sens1 = mt.reg(df_sens1, y_name = 'nonOECDshare_diff', x_name = regressors_sens1, cluster = 'kommun')
# second stage
iv_sens1 = mt.ivreg(df_sens1, y_name = 'soc_bidr_diff', x_name = 'nonOECDshare_diff', z_name = 'flyktingandel',
w_name = exog_sens1, iv_method = '2sls', cluster = 'kommun')
# call function from auxiliary file that creates LaTex table fragments for Table 7
get_table7(df, reg_sens1, iv_sens1)
return(reg_sens1, iv_sens1)
def regress(dependent_variable, dataframe, degree):
"""Regress the dependent variables on covariates (independent variables).
Args:
dependent_variable (float): the independent variable
dataframe (pd.DataFrame): the dataframe of full sample, narrow window, and wide window
degree (integer): degree of polynomials
Returns:
regression result(result)
"""
reg = mt.reg(dataframe,
f"{dependent_variable}",
get_covariates(degree),
cluster="score")
return reg
def first_stage(self, exogenous_varname, endogenous_varname, IV_varname):
'''Check the first stage for IVs
All input must be lists '''
import econtools
import econtools.metrics as mt
# get data
df = self.products.sort_values(['market_id', 'product_id']).copy()
x_varnames = exogenous_varname + IV_varname
# first stage reg
# 1. partial out x:
#for y_varname in endogenous_varname:
# result = mt.reg(df, y_varname, exogenous_varname)
# df['{}_ddot'.format(y_varname)] = df[y_varname] - df[exogenous_varname][email protected]
#IV_varname_ddot = []
#for iv_varname in IV_varname:
# result = mt.reg(df, iv_varname, exogenous_varname)
# df['{}_ddot'.format(iv_varname)] = df[iv_varname] - df[exogenous_varname][email protected]
# IV_varname_ddot = IV_varname_ddot + ['{}_ddot'.format(iv_varname)]
# 2. First stage regression
for y_varname in endogenous_varname:
#result = mt.reg(df, '{}_ddot'.format(y_varname), IV_varname_ddot)
result = mt.reg(df, y_varname, x_varnames)
joint_F = result.Ftest(IV_varname)
print('#=======================================================')
print('Endogenous var: {} '.format(y_varname))
print(' IV: {} '.format(IV_varname))
print(' ')
print('Ftest = {}'.format(joint_F))
print(' ')
print('regression:')
print(result)
print(' ')
return
def sens2_reg(df, df_sens2):
# run regression without dummy for panel period 1991/94 as we excluded the observations of that panel
# otherwise all values of that variable are zero and we cannot invert the regressor matrix
regressors_sens2 = ['const', 'mean_hvac_share', 'mean_unempl', 'munempldiff', 'mskattekrdiff', 'mbefdiff', 'ekbistandpc_diff',
'dsmastad', 'dstorstad', 'dsv_maj_1', 'dseatsmp_1', 'dseatsop_1', 'panel8891', 'flyktingandel']
exog_sens2 = ['const', 'mean_hvac_share', 'mean_unempl', 'munempldiff', 'mskattekrdiff', 'mbefdiff', 'ekbistandpc_diff', 'dsmastad',
'dstorstad', 'dsv_maj_1', 'dseatsmp_1', 'dseatsop_1', 'panel8891']
# first stage
reg_sens2 = mt.reg(df_sens2, y_name = 'nonOECDshare_diff', x_name = regressors_sens2, cluster = 'kommun')
# second stage
iv_sens2 = mt.ivreg(df_sens2, y_name = 'soc_bidr_diff', x_name = 'nonOECDshare_diff', z_name = 'flyktingandel', w_name = exog_sens2,
iv_method = '2sls', cluster = 'kommun')
# call function from auxiliary file that creates LaTex table fragments for Table 8
get_table8(df, reg_sens2, iv_sens2)
# perform F-test for relevance of the instrument
F_stat = reg_sens2.Ftest('flyktingandel')[0]
return(reg_sens2, iv_sens2, F_stat)
def _OLS_second_stage(self, df):
res = mt.reg(df, 'log_PA_PB', ['const', 'NA_NB_diff'])
return res
def _OLS_first_stage(self, df):
res = mt.reg(df, 'log_PB_P0', ['const', 'X_m', 'N_B', 'N_A'])
return res
print(df.head())
df = df.dropna()
df = df[(df != 0).all(1)]
df = df[df.year < 2017]
df = df[df.year > 1972]
print(df.head())
df['leadGDP'] = np.log(df.gdppercapitaconstantlcunygdppcap.shift(-5))
print(df.head())
#creatiing growth10 var
df['growth5'] = (df.leadGDP) - np.log(df.gdppercapitaconstantlcunygdppcap)
y = 'growth5'
X = ['Competitive_Rise_Country', 'Competitive_Decline_Country']
fe_var = 'country'
print(df.head())
results = mt.reg(
df, # DataFrame
y, # Dependent var (string)
X, # Independent var(s) (string or list of strings)
fe_name=fe_var,
addcons=True # Adds a constant term
)
print(results)
import pandas as pd import numpy as np from econtools.metrics import reg X = pd.DataFrame(np.stack([np.array([1]*10),np.array([0,1]*5),np.array([1,0]*5)]).T) beta = np.array([1,2,3]) y = X@beta data = pd.DataFrame(X) data.columns = ['cons','x1','x2'] data['y'] = y # This is fine reg(data,'y',['cons','x1']) # This should not be fine reg(data,'y',['cons','x1','x2'])
