🔬 PyCupFM¶

📦 Install in one line¶

pip install pycupfm

✨ Features at a Glance¶

📐

5 Estimators¶

LSDV, Bai FM, CupFM ★, CupFM-bar, CupBC — faithfully translated from the original GAUSS source code

🎨

Beautiful Visualizations¶

9 publication-quality plot types with premium academic aesthetics using matplotlib

🔍

Auto Factor Selection¶

Bai & Ng (2002) IC for automatic factor number selection + auto-bandwidth

📤

Export Everywhere¶

LaTeX, Excel, CSV, HTML tables — ready for your next paper submission

🧪

Monte Carlo Tools¶

Built-in DGP simulation replicating BKN (2009) Tables 1-4

🐼

Pandas Native¶

DataFrame input/output with automatic variable name inference

🚀 Quick Start — 5 Lines of Code¶

from pycupfm import CupFM
from pycupfm.datasets import load_grunfeld

df = load_grunfeld()  # N=10 firms, T=20 years
model = CupFM(n_factors=1, bandwidth=3, max_iter=25)
results = model.fit(
    y=df['linvest'], X=df[['lmvalue', 'lkstock']],
    panel_id=df['firm'], time_id=df['year'],
    var_names=['lmvalue', 'lkstock'], dep_var='linvest'
)
results.summary()

📊 Output — Publication-Quality Summary Table¶

============================================================================== cupfm — Panel Cointegration with Common Factors v1.0.0 Bai, Kao & Ng (2009, JoE 149:82-99) | Bai & Kao (2005, SSRN-1815227) ============================================================================== Panel Information -------------------------------------------------------------------------- Dependent variable : linvest Regressors : lmvalue, lkstock Cross-sections (N) : 10 Time periods (T) : 20 Observations (N×T) : 200 Panel type : Balanced Common factors (r) : 1 Bandwidth (M) : 3 (bartlett) Max iterations : 25 CupFM iterations : 25 -------------------------------------------------------------------------- Estimation Results -------------------------------------------------------------------------- Variable | LSDV Bai FM CupFM CupFM-bar CupBC --------------+---------------------------------------------------------- lmvalue | 0.5224*** 0.5840*** 0.7421*** 0.6952*** 0.5824*** | ( 6.81) ( 11.41) ( 10.43) ( 7.58) ( 11.38) --------------+---------------------------------------------------------- lkstock | 0.0827*** 0.0789*** 0.0664** 0.0716* 0.0791*** | ( 3.35) ( 2.92) ( 2.34) ( 1.87) ( 2.93) -------------------------------------------------------------------------- t-statistics in parentheses | *** p<0.01 ** p<0.05 * p<0.10 CupFM = recommended (BKN 2009, Theorem 3) | CupFM-bar = Z-bar variant

🎨 Visualization Gallery¶

📊 Coefficient Comparison — All 5 estimators with 95% confidence intervals

📈 Estimated Common Factors — F̂ₜ time series with r = 1

📉 Factor Loadings — Heterogeneous λᵢ across cross-sections

🔄 CupFM Convergence — β estimates across iterations

🌡️ Long-Run Covariance Ω̂ — Kernel-estimated covariance matrix

⭐ Bai-Ng IC — Automatic factor number selection (r* = optimal)

📐 The Model¶

Panel cointegrating regression with common factor structure:

\[y_{it} = \alpha_i + \beta' x_{it} + \lambda_i' F_t + u_{it}\]

where \(F_t\) are \(r\) common I(1) stochastic trends and \(\lambda_i\) are heterogeneous loadings.

Estimator	Method	Iterates	Reference	Status
LSDV	Within / FE	✗	Biased baseline	⚠️ Biased
Bai FM	FM correction (1-step)	✗	Bai & Kao (2005)	✅ Consistent
CupFM ★	FM + continuous updating	✓	BKN (2009) Thm 3	⭐ Recommended
CupFM-bar	FM + Z-bar instrument	✓	BKN (2009)	✅ Consistent
CupBC	BC + updating	✓	BKN (2009) Thm 2	✅ Consistent

🔗 Also Available for Stata¶

ssc install cupfm
cupfm y x1 x2, nfactors(2) bandwidth(5) plot

👤 Dr. Merwan Roudane¶

📧 merwanroudane920@gmail.com

🔗 GitHub · PyPI · Stata SSC