greeners 1.4.9

High-performance econometrics with R/Python formulas. Two-Way Clustering, Marginal Effects (AME/MEM), HC1-4, IV Predictions, Categorical C(var), Polynomial I(x^2), Interactions, Diagnostics. OLS, IV/2SLS, DiD, Logit/Probit, Panel (FE/RE), Time Series (VAR/VECM), Quantile!
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290

"""
Generate reference values from statsmodels/scipy for numerical accuracy tests in Greeners.
Output: tests/reference_values.csv (key-value format)

Requirements: pip install numpy statsmodels linearmodels
"""

import csv
import numpy as np
import statsmodels.api as sm
from statsmodels.discrete.discrete_model import Logit, Probit
from statsmodels.regression.quantile_regression import QuantReg
from linearmodels.panel import PanelOLS, RandomEffects

np.random.seed(42)

rows = []


def add(key, value):
    rows.append((key, f"{value:.10f}"))


# ============================================================================
# DATASET 1: OLS (n=50, k=3 regressors + intercept)
# ============================================================================
n = 50
x1 = np.random.randn(n)
x2 = np.random.uniform(0, 10, n)
x3 = np.random.exponential(2, n)
eps = np.random.randn(n) * 1.5
y = 3.0 + 1.5 * x1 - 0.8 * x2 + 0.5 * x3 + eps

# Save raw data so Rust tests can reconstruct the exact same dataset
for i in range(n):
    add(f"ols_data.y.{i}", y[i])
    add(f"ols_data.x1.{i}", x1[i])
    add(f"ols_data.x2.{i}", x2[i])
    add(f"ols_data.x3.{i}", x3[i])

X = sm.add_constant(np.column_stack([x1, x2, x3]))

# OLS NonRobust
res = sm.OLS(y, X).fit()
for i, name in enumerate(["const", "x1", "x2", "x3"]):
    add(f"ols_nonrobust.params.{name}", res.params[i])
    add(f"ols_nonrobust.se.{name}", res.bse[i])
    add(f"ols_nonrobust.t.{name}", res.tvalues[i])
    add(f"ols_nonrobust.p.{name}", res.pvalues[i])
add("ols_nonrobust.r_squared", res.rsquared)
add("ols_nonrobust.adj_r_squared", res.rsquared_adj)
add("ols_nonrobust.f_statistic", res.fvalue)
add("ols_nonrobust.log_likelihood", res.llf)
add("ols_nonrobust.aic", res.aic)
add("ols_nonrobust.bic", res.bic)
add("ols_nonrobust.sigma", np.sqrt(res.mse_resid))
add("ols_nonrobust.n_obs", float(res.nobs))
add("ols_nonrobust.df_resid", float(res.df_resid))

# OLS HC1
res_hc1 = sm.OLS(y, X).fit(cov_type="HC1")
for i, name in enumerate(["const", "x1", "x2", "x3"]):
    add(f"ols_hc1.se.{name}", res_hc1.bse[i])
    add(f"ols_hc1.t.{name}", res_hc1.tvalues[i])
    add(f"ols_hc1.p.{name}", res_hc1.pvalues[i])

# OLS HC3
res_hc3 = sm.OLS(y, X).fit(cov_type="HC3")
for i, name in enumerate(["const", "x1", "x2", "x3"]):
    add(f"ols_hc3.se.{name}", res_hc3.bse[i])
    add(f"ols_hc3.t.{name}", res_hc3.tvalues[i])
    add(f"ols_hc3.p.{name}", res_hc3.pvalues[i])

# OLS with Normal inference (z-values instead of t-values)
res_norm = sm.OLS(y, X).fit(use_t=False)
for i, name in enumerate(["const", "x1", "x2", "x3"]):
    add(f"ols_normal.z.{name}", res_norm.tvalues[i])
    add(f"ols_normal.p.{name}", res_norm.pvalues[i])

# OLS HC2
res_hc2 = sm.OLS(y, X).fit(cov_type="HC2")
for i, name in enumerate(["const", "x1", "x2", "x3"]):
    add(f"ols_hc2.se.{name}", res_hc2.bse[i])
    add(f"ols_hc2.t.{name}", res_hc2.tvalues[i])
    add(f"ols_hc2.p.{name}", res_hc2.pvalues[i])

# Note: HC4 is not supported by statsmodels, so no reference values for it.

# OLS Newey-West (HAC)
res_nw = sm.OLS(y, X).fit(cov_type="HAC", cov_kwds={"maxlags": 3})
for i, name in enumerate(["const", "x1", "x2", "x3"]):
    add(f"ols_nw.se.{name}", res_nw.bse[i])
    add(f"ols_nw.t.{name}", res_nw.tvalues[i])
    add(f"ols_nw.p.{name}", res_nw.pvalues[i])

# OLS Clustered (create cluster IDs: 10 clusters of 5 obs each)
cluster_ids = np.repeat(np.arange(10), 5)
res_cl = sm.OLS(y, X).fit(cov_type="cluster", cov_kwds={"groups": cluster_ids})
for i, name in enumerate(["const", "x1", "x2", "x3"]):
    add(f"ols_clustered.se.{name}", res_cl.bse[i])
    add(f"ols_clustered.t.{name}", res_cl.tvalues[i])
    add(f"ols_clustered.p.{name}", res_cl.pvalues[i])
# Save cluster IDs for Rust test
for i in range(n):
    add(f"ols_data.cluster.{i}", float(cluster_ids[i]))

# ============================================================================
# DATASET 2: IV / 2SLS (n=100)
# ============================================================================
n = 100
z1 = np.random.randn(n)
z2 = np.random.randn(n)
u = np.random.randn(n)
x_endog = 0.5 * z1 + 0.3 * z2 + 0.5 * u  # endogenous regressor
y_iv = 2.0 + 1.0 * x_endog + u

for i in range(n):
    add(f"iv_data.y.{i}", y_iv[i])
    add(f"iv_data.x_endog.{i}", x_endog[i])
    add(f"iv_data.z1.{i}", z1[i])
    add(f"iv_data.z2.{i}", z2[i])

from statsmodels.sandbox.regression.gmm import IV2SLS

res_iv = IV2SLS(
    y_iv,
    sm.add_constant(x_endog),
    sm.add_constant(np.column_stack([z1, z2])),
).fit()
for i, name in enumerate(["const", "x_endog"]):
    add(f"iv_2sls.params.{name}", res_iv.params[i])
    add(f"iv_2sls.se.{name}", res_iv.bse[i])
add("iv_2sls.r_squared", res_iv.rsquared)

# ============================================================================
# DATASET 3: Logit & Probit (n=200)
# ============================================================================
n = 200
x1_bin = np.random.randn(n)
x2_bin = np.random.uniform(-2, 2, n)
latent = -1.0 + 0.8 * x1_bin + 1.2 * x2_bin + np.random.logistic(0, 1, n)
y_bin = (latent > 0).astype(float)

for i in range(n):
    add(f"binary_data.y.{i}", y_bin[i])
    add(f"binary_data.x1.{i}", x1_bin[i])
    add(f"binary_data.x2.{i}", x2_bin[i])

X_bin = sm.add_constant(np.column_stack([x1_bin, x2_bin]))

# Logit
res_logit = Logit(y_bin, X_bin).fit(disp=0)
for i, name in enumerate(["const", "x1", "x2"]):
    add(f"logit.params.{name}", res_logit.params[i])
    add(f"logit.se.{name}", res_logit.bse[i])
    add(f"logit.z.{name}", res_logit.tvalues[i])
    add(f"logit.p.{name}", res_logit.pvalues[i])
add("logit.log_likelihood", res_logit.llf)
add("logit.pseudo_r2", res_logit.prsquared)

# Probit
res_probit = Probit(y_bin, X_bin).fit(disp=0)
for i, name in enumerate(["const", "x1", "x2"]):
    add(f"probit.params.{name}", res_probit.params[i])
    add(f"probit.se.{name}", res_probit.bse[i])
    add(f"probit.z.{name}", res_probit.tvalues[i])
    add(f"probit.p.{name}", res_probit.pvalues[i])
add("probit.log_likelihood", res_probit.llf)
add("probit.pseudo_r2", res_probit.prsquared)

# Logit: average marginal effects
mfx = res_logit.get_margeff(at="overall")
for i, name in enumerate(["x1", "x2"]):
    add(f"logit.ame.{name}", mfx.margeff[i])
    add(f"logit.ame_se.{name}", mfx.margeff_se[i])

# ============================================================================
# DATASET 4: Panel Data - Fixed & Random Effects (N=20, T=5)
# ============================================================================
N_panel, T_panel = 20, 5
n_panel = N_panel * T_panel
entity_ids = np.repeat(np.arange(N_panel), T_panel)
time_ids = np.tile(np.arange(T_panel), N_panel)
alpha_i = np.repeat(np.random.randn(N_panel) * 2, T_panel)  # entity effects
x1_panel = np.random.randn(n_panel)
x2_panel = np.random.uniform(0, 5, n_panel)
eps_panel = np.random.randn(n_panel) * 0.5
y_panel = 1.0 + 2.0 * x1_panel - 0.5 * x2_panel + alpha_i + eps_panel

for i in range(n_panel):
    add(f"panel_data.y.{i}", y_panel[i])
    add(f"panel_data.x1.{i}", x1_panel[i])
    add(f"panel_data.x2.{i}", x2_panel[i])
    add(f"panel_data.entity.{i}", float(entity_ids[i]))
    add(f"panel_data.time.{i}", float(time_ids[i]))

import pandas as pd

panel_df = pd.DataFrame(
    {
        "y": y_panel,
        "x1": x1_panel,
        "x2": x2_panel,
        "entity": entity_ids,
        "time": time_ids,
    }
)
panel_df = panel_df.set_index(["entity", "time"])

# Fixed Effects (within estimator)
fe_mod = PanelOLS(panel_df["y"], panel_df[["x1", "x2"]], entity_effects=True)
fe_res = fe_mod.fit()
for i, name in enumerate(["x1", "x2"]):
    add(f"panel_fe.params.{name}", fe_res.params[name])
    add(f"panel_fe.se.{name}", fe_res.std_errors[name])
    add(f"panel_fe.t.{name}", fe_res.tstats[name])
    add(f"panel_fe.p.{name}", fe_res.pvalues[name])
add("panel_fe.r_squared", fe_res.rsquared)
add("panel_fe.n_obs", float(fe_res.nobs))

# Random Effects
re_mod = RandomEffects(panel_df["y"], sm.add_constant(panel_df[["x1", "x2"]]))
re_res = re_mod.fit()
for name in ["const", "x1", "x2"]:
    add(f"panel_re.params.{name}", re_res.params[name])
    add(f"panel_re.se.{name}", re_res.std_errors[name])
    add(f"panel_re.t.{name}", re_res.tstats[name])
    add(f"panel_re.p.{name}", re_res.pvalues[name])
add("panel_re.r_squared", re_res.rsquared)

# ============================================================================
# DATASET 5: Quantile Regression (n=100, tau=0.25, 0.50, 0.75)
# ============================================================================
n = 100
x1_q = np.random.randn(n)
x2_q = np.random.uniform(0, 5, n)
# Heteroskedastic errors to make quantiles differ
eps_q = np.random.randn(n) * (1 + 0.5 * np.abs(x1_q))
y_q = 2.0 + 1.0 * x1_q - 0.3 * x2_q + eps_q

for i in range(n):
    add(f"quantile_data.y.{i}", y_q[i])
    add(f"quantile_data.x1.{i}", x1_q[i])
    add(f"quantile_data.x2.{i}", x2_q[i])

X_q = sm.add_constant(np.column_stack([x1_q, x2_q]))

for tau in [0.25, 0.50, 0.75]:
    res_qr = QuantReg(y_q, X_q).fit(q=tau)
    tau_str = f"{tau:.2f}"
    for i, name in enumerate(["const", "x1", "x2"]):
        add(f"quantile_{tau_str}.params.{name}", res_qr.params[i])
        add(f"quantile_{tau_str}.se.{name}", res_qr.bse[i])
        add(f"quantile_{tau_str}.t.{name}", res_qr.tvalues[i])
        add(f"quantile_{tau_str}.p.{name}", res_qr.pvalues[i])

# ============================================================================
# DATASET 6: WLS (n=50, known heteroskedastic weights)
# ============================================================================
n = 50
x1_w = np.random.randn(n)
# Variance proportional to |x1| + 1
sigma_i = np.sqrt(np.abs(x1_w) + 1.0)
eps_w = np.random.randn(n) * sigma_i
y_w = 5.0 - 2.0 * x1_w + eps_w
weights = 1.0 / (sigma_i ** 2)

for i in range(n):
    add(f"wls_data.y.{i}", y_w[i])
    add(f"wls_data.x1.{i}", x1_w[i])
    add(f"wls_data.weights.{i}", weights[i])

X_w = sm.add_constant(x1_w)
res_wls = sm.WLS(y_w, X_w, weights=weights).fit()
for i, name in enumerate(["const", "x1"]):
    add(f"wls.params.{name}", res_wls.params[i])
    add(f"wls.se.{name}", res_wls.bse[i])
    add(f"wls.t.{name}", res_wls.tvalues[i])
    add(f"wls.p.{name}", res_wls.pvalues[i])
add("wls.r_squared", res_wls.rsquared)

# ============================================================================
# Write CSV
# ============================================================================
with open("tests/reference_values.csv", "w", newline="") as f:
    writer = csv.writer(f)
    writer.writerow(["key", "value"])
    writer.writerows(rows)

print(f"Written {len(rows)} reference values to tests/reference_values.csv")