pycupfm 1.0.0

Panel Cointegration with Common Factors — CupFM, CupBC, Bai FM, LSDV estimators (Bai, Kao & Ng 2009)

PyCupFM — Panel Cointegration with Common Factors

Python implementation of all 5 panel cointegration estimators from
Bai, Kao & Ng (2009, Journal of Econometrics) and Bai & Kao (2005, SSRN)

---

✨ Features

Feature	Description
5 Estimators	LSDV, Bai FM, CupFM ★, CupFM-bar, CupBC
Auto Factor Selection	Bai & Ng (2002) information criterion
3 Kernels	Bartlett, Parzen, Quadratic Spectral
Auto Bandwidth	Newey-West and Andrews plug-in rules
Beautiful Plots	9 publication-quality visualization types
Export	LaTeX, Excel, CSV, HTML tables
Monte Carlo	Built-in DGP simulation (BKN 2009 design)
Pandas Native	DataFrame input/output

📦 Installation

pip install pycupfm

🚀 Quick Start

from pycupfm import CupFM
from pycupfm.datasets import load_grunfeld

# Load classic Grunfeld investment data (N=10 firms, T=20 years)
df = load_grunfeld()

# Fit all 5 estimators
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'
)

# Publication-quality summary table
results.summary()

# Beautiful visualizations
model.plot(kind='all')

📊 Output Example

==============================================================================
  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)
  --------------------------------------------------------------------------

  Estimation Results
  --------------------------------------------------------------------------
      Variable  |       LSDV     Bai FM      CupFM    CupFM-bar      CupBC
  --------------+----------------------------------------------------------
       lmvalue  |  0.1234***  0.1189***  0.1156***  0.1180***  0.1145***
                |   (  5.67)   (  5.23)   (  5.01)   (  5.12)   (  4.95)
       lkstock  |  0.3456***  0.3312***  0.3289***  0.3301***  0.3278***
                |   (  8.91)   (  8.34)   (  8.22)   (  8.27)   (  8.15)
  --------------------------------------------------------------------------

📐 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, $\lambda_i$ are heterogeneous factor loadings, and $\beta$ is the homogeneous cointegrating vector.

Estimators

Estimator	Method	Iterates	Reference
LSDV	Within/FE	No	Biased baseline
Bai FM	FM correction (1-step)	No	Bai & Kao (2005)
CupFM ★	FM + continuous updating	Yes	BKN (2009) Thm 3
CupFM-bar	FM + Z-bar instrument	Yes	BKN (2009)
CupBC	Bias-corrected + updating	Yes	BKN (2009) Thm 2

🎨 Visualization Gallery

# Individual plots
model.plot(kind='coefficients')   # Forest plot with 95% CIs
model.plot(kind='factors')        # Estimated common factors
model.plot(kind='loadings')       # Factor loadings scatter/bar
model.plot(kind='convergence')    # CupFM iteration path
model.plot(kind='omega')          # Long-run covariance heatmap
model.plot(kind='ic')             # Bai-Ng IC values vs r

🔬 Monte Carlo Simulation

from pycupfm import simulate_panel, monte_carlo

# Simulate BKN (2009) DGP
sim = simulate_panel(N=20, T=40, k=1, r=2, beta=2.0, seed=42)

# Run Monte Carlo experiment
mc = monte_carlo(n_reps=100, N=20, T=40, verbose=True)

📤 Export Results

# Export to multiple formats
results.to_latex()           # LaTeX table
results.to_excel('out.xlsx') # Excel workbook
results.to_csv('out.csv')    # CSV file
results.to_dataframe()       # Pandas DataFrame

# Or all at once:
from pycupfm import export_results
export_results(results, 'cupfm_results', fmt='all')

🔗 Also Available for Stata

This package is the Python companion to the Stata cupfm package:

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

📚 References

• Bai, J., Kao, C. & Ng, S. (2009). Panel cointegration with global stochastic trends. Journal of Econometrics, 149(1), 82-99. DOI
• Bai, J. & Kao, C. (2005). On the estimation and inference of a panel cointegration model with cross-sectional dependence. CPR Working Paper No. 75. SSRN
• Bai, J. & Ng, S. (2002). Determining the number of factors in approximate factor models. Econometrica, 70(1), 191-221.

👤 Author

Dr. Merwan Roudane

• Email: [e-mail]
• GitHub: @merwanroudane

📄 License

MIT License — see LICENSE for details.