robcointeg 0.0.2

Outlier-Robust Cointegration Analysis based on Franses & Lucas (1998)

robcointeg

Outlier-Robust Cointegration Analysis

A Python implementation of outlier-robust cointegration testing based on the methodology of Franses & Lucas (1998).

Author: Dr Merwan Roudane
Email: [e-mail]
GitHub: https://github.com/merwanroudane/robcointeg

---

Overview

Standard unit-root and cointegration tests are sensitive to atypical events such as outliers and structural breaks. This package implements the outlier-robust cointegration test proposed by Franses & Lucas (1998), which provides:

1. Robust Cointegration Testing: A pseudolikelihood ratio (PLR) test based on the Student-t distribution that is less sensitive to outliers than the standard Gaussian-based Johansen test.

2. Outlier Detection: Observation weights that identify which data points may be driving the cointegration results.

3. Diagnostic Tools: Methods for comparing robust and non-robust test results to signal when standard cointegration results might be influenced by aberrant observations.

---

Installation

From PyPI (when published)

pip install robcointeg

From Source

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

With Plotting Support

pip install robcointeg[plotting]

Full Installation (all optional dependencies)

pip install robcointeg[full]

---

Quick Start

Basic Usage

import numpy as np
from robcointeg import RobustVAR, compare_tests, diagnostic_report

# Generate sample data
np.random.seed(42)
T = 100
y = np.cumsum(np.random.randn(T, 2), axis=0)

# Add an outlier
y[50, 0] += 5

# Fit robust VAR model
model = RobustVAR(p=2, nu=5)
model.fit(y)

# View results
print(model.summary())
print(f"Observation weights: {model.weights}")

Cointegration Testing

from robcointeg import plr_test, johansen_trace_test

# Robust test (Student-t based)
robust_result = plr_test(y, p=2, nu=5, r=0)
print(f"Robust PLR statistic: {robust_result.test_statistic:.2f}")
print(f"Reject at 5%: {robust_result.reject[0.05]}")

# Standard Johansen test for comparison
johansen_result = johansen_trace_test(y, p=2, r=0)
print(f"Johansen statistic: {johansen_result.test_statistic:.2f}")

Compare Robust and Non-Robust Tests

from robcointeg import compare_tests

comparison = compare_tests(y, p=2, nu=5)
print(comparison['summary'])

if comparison['conflict']:
    print("⚠ Tests give different results!")
    print(f"Gaussian rank: {comparison['gaussian_rank']}")
    print(f"Robust rank: {comparison['robust_rank']}")

Full Diagnostic Report

from robcointeg import diagnostic_report

report = diagnostic_report(
    y, p=2, nu=5,
    variable_names=['Y1', 'Y2'],
    significance_level=0.05
)

---

Methodology

This package implements the methodology from:

Franses, P.H. & Lucas, A. (1998). "Outlier Detection in Cointegration Analysis", Journal of Business & Economic Statistics, 16:4, 459-468.

Key Equations

1. VAR/VECM Model (Equation 1):

$$\Delta y_t = \alpha\beta' y_{t-1} + \Phi_1 \Delta y_{t-1} + \cdots + \Phi_{p-1} \Delta y_{t-p+1} + \mu + \varepsilon_t$$

2. Student-t Pseudolikelihood (Equation 2):

$$\mathcal{L}(\theta) = \prod_{t=1}^{T} \frac{\Gamma((\nu+k)/2)}{\Gamma(\nu/2)|\pi\nu V|^{1/2}} \left(1 + \frac{\varepsilon_t' V^{-1} \varepsilon_t}{\nu}\right)^{-(\nu+k)/2}$$

3. Pseudolikelihood Ratio Test (Equation 3):

$$\text{PLR} = 2 \ln\left(\frac{\mathcal{L}(\hat{\theta})}{\mathcal{L}(\tilde{\theta})}\right)$$

4. Observation Weights (Equation 8):

$$w_t = \left(\frac{\nu}{\nu + \varepsilon_t' V^{-1} \varepsilon_t}\right)^{1/2}$$

Interpretation of Weights

• Low weight ($w_t < 0.67$ for $\nu=5$, $k=2$): The observation does not correspond to the general pattern of the model and deserves closer inspection.
• Consecutive low weights may indicate additive outliers (AO).
• Low weights at the start and end of a period may indicate level shifts.
• Many consecutive low weights may indicate variance shifts.

---

API Reference

Core Classes

RobustVAR

from robcointeg import RobustVAR

model = RobustVAR(p=2, nu=5, with_constant=True)
model.fit(y)

# Attributes
model.weights        # Observation weights
model.residuals      # Model residuals
model.log_likelihood # Log pseudolikelihood
model.Pi             # Long-run impact matrix
model.alpha          # Loading matrix
model.beta           # Cointegrating vectors
model.eigenvalues    # Eigenvalues

RobustCointegrationTest

from robcointeg import sequential_plr_test

result = sequential_plr_test(y, p=2, nu=5)

# Attributes
result.test_results    # List of test results for each rank
result.selected_rank   # Determined cointegrating rank
result.weights         # Observation weights
result.outlier_indices # Detected outlier indices
result.summary         # Text summary

Main Functions

Function	Description
plr_test()	Perform PLR test for a specific rank
johansen_trace_test()	Standard Johansen trace test
compare_tests()	Compare robust and non-robust tests
compute_weights()	Compute observation weights
detect_outliers()	Identify outlying observations
diagnostic_report()	Generate full diagnostic report

Critical Values

The package includes critical values from Table 1 of Franses & Lucas (1998):

from robcointeg import get_critical_value, CriticalValueTable

# Get single critical value
cv = get_critical_value(k_minus_r=2, nu=5, significance_level=0.05)

# Access full table
table = CriticalValueTable(with_drift=True)
print(table.print_table())

---

Examples

Example 1: Finland/U.S. Real Exchange Rate

This replicates the empirical application from Section 3 of the paper:

import numpy as np
from robcointeg import compare_tests, plot_comparison_panel

# Load Finland/U.S. exchange rate data (hypothetical)
# y should contain: log(CPI_US), log(CPI_Finland), log(Exchange_Rate)

comparison = compare_tests(y, p=2, nu=5)
print(comparison['summary'])

# Visualize (requires matplotlib)
plot_comparison_panel(
    y, comparison['weights'], comparison['outlier_indices'],
    nu=5, k=3, title='Finland/U.S. Real Exchange Rate Analysis'
)

Example 2: Monte Carlo Simulation

from robcointeg import monte_carlo_critical_values

# Simulate critical values
cv = monte_carlo_critical_values(
    k_minus_r=2, nu=5, n_sim=10000, T=100, with_drift=True
)
print(f"95% critical value: {cv[0.95]:.2f}")

---

Citation

If you use this package in your research, please cite the original paper:

@article{franses1998outlier,
  title={Outlier Detection in Cointegration Analysis},
  author={Franses, Philip Hans and Lucas, Andr{\'e}},
  journal={Journal of Business \& Economic Statistics},
  volume={16},
  number={4},
  pages={459--468},
  year={1998},
  publisher={Taylor \& Francis}
}

And this software:

@software{robcointeg,
  author = {Roudane, Merwan},
  title = {robcointeg: Outlier-Robust Cointegration Analysis in Python},
  year = {2025},
  url = {https://github.com/merwanroudane/robcointeg}
}

---

References

• Franses, P.H. & Lucas, A. (1998). "Outlier Detection in Cointegration Analysis", Journal of Business & Economic Statistics, 16:4, 459-468.
• Johansen, S. (1988). "Statistical Analysis of Cointegration Vectors", Journal of Economic Dynamics and Control, 12, 231-254.
• Johansen, S. (1991). "Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models", Econometrica, 59, 1551-1580.
• Lucas, A. (1997). "Cointegration Testing Using Pseudo Likelihood Ratio Tests", Econometric Theory, 13, 149-169.
• Lucas, A. (1998). "Inference on Cointegrating Ranks Using LR and LM Tests Based on Pseudo Likelihoods", Econometric Reviews, 17, 185-214.

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

---

Changelog

Version 0.0.1 (2025)

• Initial release
• Implementation of Franses & Lucas (1998) methodology
• Student-t pseudolikelihood ratio (PLR) test
• Observation weight computation
• Outlier detection
• Comparison with standard Johansen test
• Visualization tools
• Critical value tables from the original paper