pydcce

Dynamic Common Correlated Effects Estimation for Panel Data with Cross-Sectional Dependence

A comprehensive Python implementation of Jan Ditzen's Stata xtdcce2 package for estimating heterogeneous coefficient panel data models with cross-sectional dependence.

Author: Dr. Merwan Roudane
Email: merwanroudane920@gmail.com
GitHub: https://github.com/merwanroudane/pydecce2

Table of Contents

1. Installation
2. Features
3. Quick Start
4. Estimators
   • Mean Group (MG)
   • Common Correlated Effects (CCE)
   • Dynamic CCE
   • Pooled CCE
   • Error Correction Model (ECM/PMG)
   • CS-DL
   • CS-ARDL
   • Regularized CCE (rCCE)
5. Cross-Sectional Dependence Tests
6. Exponent Estimation
7. API Reference
8. References

Installation

pip install pydcce

Or install from source:

git clone https://github.com/merwanroudane/pydecce2.git
cd pydecce2
pip install -e .

Dependencies

• numpy >= 1.20.0
• pandas >= 1.3.0
• scipy >= 1.7.0
• statsmodels >= 0.13.0
• tabulate >= 0.9.0
• matplotlib >= 3.5.0
• seaborn >= 0.11.0

Features

Estimators

• Mean Group (MG) - Pesaran & Smith (1995)
• CCE - Common Correlated Effects (Pesaran, 2006)
• Dynamic CCE - Chudik & Pesaran (2015)
• Pooled CCE - Homogeneous coefficients with CSD
• ECM/PMG - Error Correction Model
• CS-DL - Cross-Section Augmented Distributed Lag
• CS-ARDL - Cross-Section Augmented ARDL
• rCCE - Regularized CCE (Juodis, 2022)

Tests

• CD Test - Pesaran (2015, 2021)
• CDw - Weighted CD (Juodis & Reese, 2021)
• CDw+ - Power Enhanced CDw
• CD* - Pesaran & Xie (2021)
• Alpha Estimation - Bailey, Kapetanios, Pesaran (2016)

Output

• Beautiful tables using tabulate
• Stata-like output format
• LaTeX and HTML export
• Visualizations (heatplots, density plots)

Quick Start

import pandas as pd
from pydcce import (
    CCE, DynamicCCE, MeanGroup,
    CDTest, ExponentEstimator, PanelData
)

# Load panel data (long format with 'id' and 'year' columns)
df = pd.read_excel('mydata.xlsx')

# Check panel structure
panel = PanelData(df, unit_col='id', time_col='year')
print(panel.summary())

# Test for cross-sectional dependence
cd = CDTest(df, var='GDP', unit_col='id', time_col='year')
cd_result = cd.test(pesaran=True, cdw=True, pea=True, cdstar=True)
print(cd_result)

# Prepare model data with lagged dependent variable
df_model = df.copy().sort_values(['id', 'year'])
df_model['L_GDP'] = df_model.groupby('id')['GDP'].shift(1)
df_model = df_model.dropna()

# Mean Group estimation
mg = MeanGroup(
    data=df_model,
    depvar='GDP',
    indepvars=['L_GDP', 'INF', 'UR', 'EXP', 'FDI'],
    unit_col='id',
    time_col='year',
    constant=True
)
mg_result = mg.fit()
print(mg_result.summary_table())

# CCE estimation (accounts for cross-sectional dependence)
cce = CCE(
    data=df_model,
    depvar='GDP',
    indepvars=['L_GDP', 'INF', 'UR', 'EXP', 'FDI'],
    unit_col='id',
    time_col='year',
    constant=True
)
cce_result = cce.fit()
print(cce_result.summary_table())

# Dynamic CCE with CSA lags (Chudik & Pesaran, 2015)
dcce = DynamicCCE(
    data=df_model,
    depvar='GDP',
    indepvars=['L_GDP', 'INF', 'UR', 'EXP', 'FDI'],
    unit_col='id',
    time_col='year',
    csa_lags=2,
    constant=True
)
dcce_result = dcce.fit()
print(dcce_result.summary_table())

Estimators

Mean Group (MG)

Pesaran & Smith (1995) Mean Group Estimator for heterogeneous panels without cross-sectional dependence.

from pydcce import MeanGroup

mg = MeanGroup(
    data=panel_data,
    depvar='y',
    indepvars=['x1', 'x2'],
    unit_col='id',
    time_col='time',
    constant=True
)
result = mg.fit()
print(result)

# Get individual coefficients
individual_coefs = mg.get_individual_coefficients()
print(individual_coefs)

Output:

╔══════════════════════════════════════════════════════════════════════════════╗
║                          Mean Group Estimation Results                        ║
╠══════════════════════════════════════════════════════════════════════════════╣
║  Dependent Variable: y                                                        ║
║  Estimator:          Mean Group                                               ║
╠══════════════════════════════════════════════════════════════════════════════╣
║  N (units):     50               T (periods):    30                           ║
║  Observations:  1500             DF Residual:    49                           ║
║  R-squared:     0.856432         Adj. R-squared: 0.845123                     ║
╚══════════════════════════════════════════════════════════════════════════════╝

╒══════════════╤═══════════════╤════════════╤══════════╤═════════╤═══════════════════════╕
│ Variable     │   Coefficient │ Std. Error │   t-stat │   P>|t| │ 95% Conf. Int.        │
╞══════════════╪═══════════════╪════════════╪══════════╪═════════╪═══════════════════════╡
│ _cons        │      0.234567 │   0.012345 │  19.0012 │  0.0000 │ [0.2102, 0.2589]      │
│ x1           │      0.456789 │   0.023456 │  19.4723 │  0.0000 │ [0.4108, 0.5028]      │
│ x2           │      0.312456 │   0.034567 │   9.0398 │  0.0000 │ [0.2447, 0.3802]      │
╘══════════════╧═══════════════╧════════════╧══════════╧═════════╧═══════════════════════╛

Common Correlated Effects (CCE)

Pesaran (2006) CCE estimator accounting for cross-sectional dependence.

from pydcce import CCE

cce = CCE(
    data=panel_data,
    depvar='log_gdp',
    indepvars=['log_capital', 'log_labor'],
    unit_col='country',
    time_col='year',
    csa_vars=['log_gdp', 'log_capital', 'log_labor'],  # Variables for CSA
    constant=True
)
result = cce.fit()
print(result)

Dynamic CCE

Chudik & Pesaran (2015) Dynamic CCE with lagged cross-sectional averages.

from pydcce import DynamicCCE

# Add lagged dependent variable
data['L_log_gdp'] = data.groupby('country')['log_gdp'].shift(1)

dcce = DynamicCCE(
    data=data,
    depvar='log_gdp',
    indepvars=['L_log_gdp', 'log_capital', 'log_labor'],
    unit_col='country',
    time_col='year',
    csa_lags=2,  # Number of CSA lags (p_T)
    csa_vars=['log_gdp', 'log_capital', 'log_labor']
)
result = dcce.fit()
print(result)

Pooled CCE

CCE with homogeneous (pooled) coefficients.

from pydcce import PooledCCE

pcce = PooledCCE(
    data=panel_data,
    depvar='log_gdp',
    indepvars=['log_capital', 'log_labor'],
    unit_col='country',
    time_col='year',
    csa_lags=0
)
result = pcce.fit()
print(result)

Error Correction Model (ECM/PMG)

Shin et al. (1999) with long-run and short-run coefficients.

from pydcce import ECM

# Prepare differenced variables
data['D_log_gdp'] = data.groupby('country')['log_gdp'].diff()
data['D_log_capital'] = data.groupby('country')['log_capital'].diff()
data['L_log_gdp'] = data.groupby('country')['log_gdp'].shift(1)

ecm = ECM(
    data=data,
    depvar='D_log_gdp',
    lr_vars=['L_log_gdp', 'log_capital'],  # Long-run variables
    sr_vars=['D_log_capital'],             # Short-run variables
    unit_col='country',
    time_col='year'
)
result = ecm.fit()
print(result)

CS-DL

Cross-Section Augmented Distributed Lag (Chudik et al., 2016).

from pydcce import CSDL

csdl = CSDL(
    data=panel_data,
    depvar='log_gdp',
    lr_vars=['log_capital', 'log_labor'],
    px=2,  # Lags of differences
    unit_col='country',
    time_col='year',
    csa_lags=2
)
result = csdl.fit()
print(result)

CS-ARDL

Cross-Section Augmented ARDL (Chudik et al., 2016).

from pydcce import CSARDL

csardl = CSARDL(
    data=panel_data,
    depvar='log_gdp',
    indepvars=['log_capital', 'log_labor'],
    py=1,  # Lags of dependent variable
    px=1,  # Lags of independent variables
    unit_col='country',
    time_col='year',
    csa_lags=2
)
result = csardl.fit()
print(result)

# Long-run coefficients are automatically computed
# Look for LR_log_capital, LR_log_labor in output

Regularized CCE (rCCE)

Juodis (2022) regularized CCE with automatic factor selection.

from pydcce import RCCE

rcce = RCCE(
    data=panel_data,
    depvar='log_gdp',
    indepvars=['log_capital', 'log_labor'],
    unit_col='country',
    time_col='year',
    n_factors='auto',  # Automatic ER criterion
    bootstrap_reps=100  # Bootstrap SE
)
result = rcce.fit()
print(result)

Cross-Sectional Dependence Tests

Test for weak cross-sectional dependence in residuals.

from pydcce import CDTest

# Test residuals
cd = CDTest(
    data=residuals_data,
    var='residuals',
    unit_col='country',
    time_col='year'
)

result = cd.test(
    pesaran=True,   # Pesaran CD
    cdw=True,       # Weighted CD
    pea=True,       # Power enhanced
    cdstar=True,    # CD*
    cdw_reps=30,    # Replications for CDw
    n_pca=4         # Factors for CD*
)
print(result)

Output:

╔══════════════════════════════════════════════════════════════════════════════╗
║           Testing for Weak Cross-Sectional Dependence (CSD)                  ║
╠══════════════════════════════════════════════════════════════════════════════╣
║  H0: Weak cross-sectional dependence                                         ║
║  H1: Strong cross-sectional dependence                                        ║
╠══════════════════════════════════════════════════════════════════════════════╣
║  N (units):     50               T (periods):    30                           ║
╚══════════════════════════════════════════════════════════════════════════════╝

╒═══════════════════════╤═════════════╤══════════╕
│ Test                  │   Statistic │  P-value │
╞═══════════════════════╪═════════════╪══════════╡
│ CD (Pesaran)          │     12.3456 │   0.0000 │
│ CDw (Juodis & Reese)  │      8.7654 │   0.0000 │
│ CDw+ (Power Enhanced) │      9.1234 │   0.0000 │
│ CD* (Pesaran & Xie)   │      6.5432 │   0.0000 │
╘═══════════════════════╧═════════════╧══════════╛

Mean ρ(i,j) = 0.2345

Exponent Estimation

Estimate the exponent of cross-sectional dependence (α).

from pydcce import ExponentEstimator

exp_est = ExponentEstimator(
    data=panel_data,
    var='residuals',
    unit_col='country',
    time_col='year',
    n_pca=4
)

result = exp_est.estimate(
    size=0.1,
    bootstrap_reps=100
)
print(result)

Output:

╔══════════════════════════════════════════════════════════════════════════════╗
║         Cross-Sectional Dependence Exponent Estimation                       ║
╠══════════════════════════════════════════════════════════════════════════════╣
║  N (units):     50               T (periods):    30                           ║
╚══════════════════════════════════════════════════════════════════════════════╝

╒═══════════╤════════════╤════════════╤═════════════════════════╕
│ Parameter │   Estimate │ Std. Error │ 95% Conf. Int.          │
╞═══════════╪════════════╪════════════╪═════════════════════════╡
│ Alpha     │   0.678234 │   0.045678 │ [0.5887, 0.7678]        │
╘═══════════╧════════════╧════════════╧═════════════════════════╛

Interpretation:
  • alpha < 0.5   → Weak cross-sectional dependence
  • alpha >= 0.5  → Strong cross-sectional dependence
  
  Current estimate: alpha = 0.6782
  Status: STRONG dependence

API Reference

Panel Data

from pydcce import PanelData

panel = PanelData(
    data=df,
    unit_col='country',
    time_col='year'
)

# Properties
print(panel.N)           # Number of units
print(panel.T)           # Time periods
print(panel.is_balanced) # Balance status

# Methods
panel.add_lag('y', lags=2)
panel.add_difference('y', order=1)
panel.add_cross_sectional_mean(['y', 'x'], lags=2)
print(panel.summary())

All Estimators

All estimators follow the same interface:

estimator = CCE(data, depvar, indepvars, unit_col, time_col, **kwargs)
result = estimator.fit()

# Result attributes
result.coefficients     # Dict of coefficients
result.std_errors      # Dict of standard errors
result.t_stats         # Dict of t-statistics
result.p_values        # Dict of p-values
result.conf_int        # Dict of confidence intervals
result.individual_coefs # DataFrame of unit-specific coefs
result.residuals       # Array of residuals
result.r_squared       # R-squared
result.N, result.T     # Panel dimensions

# Methods
print(result.summary_table())
df = result.to_dataframe()

References

1. Pesaran, M.H. (2006). Estimation and inference in large heterogeneous panels with a multifactor error structure. Econometrica, 74(4), 967-1012.

2. Chudik, A., & Pesaran, M.H. (2015). Common correlated effects estimation of heterogeneous dynamic panel data models with weakly exogenous regressors. Journal of Econometrics, 188(2), 393-420.

3. Ditzen, J. (2018). Estimating dynamic common-correlated effects in Stata. The Stata Journal, 18(3), 585-617.

4. Bailey, N., Kapetanios, G., & Pesaran, M.H. (2016). Exponent of cross-sectional dependence: Estimation and inference. Journal of Applied Econometrics, 31(6), 929-960.

5. Juodis, A., & Reese, S. (2021). The incidental parameters problem in testing for remaining cross-section correlation. Econometric Reviews.

6. Pesaran, M.H. (2015). Testing weak cross-sectional dependence in large panels. Econometric Reviews, 34(6-10), 1089-1117.

7. Pesaran, M.H., & Xie, Y. (2021). A bias-corrected CD test for error cross-sectional dependence in panel data models. Empirical Economics.

License

MIT License

Citation

If you use pydcce in your research, please cite:

@software{pydcce,
  author = {Roudane, Merwan},
  title = {pydcce: Dynamic Common Correlated Effects for Python},
  year = {2025},
  url = {https://github.com/merwanroudane/pydecce2}
}

Contributing

Contributions are welcome! Please submit issues and pull requests on GitHub.

Developed by Dr. Merwan Roudane