API Reference¶

Main Class¶

`CupFM`¶

from pycupfm import CupFM

model = CupFM(
    n_factors=2,          # int or 'auto'
    kernel='bartlett',    # 'bartlett', 'parzen', 'qs'
    bandwidth=5,          # int or 'auto'
    max_iter=20,          # int (1-500)
    tol=1e-4,             # float
    auto_rmax=8,          # int (max r for auto-selection)
    verbose=False,        # bool
)

`model.fit(y, X, panel_id, time_id, var_names=None, dep_var='y')`¶

Fit all 5 panel cointegration estimators.

Parameters:

Parameter	Type	Description
`y`	array-like (N*T,)	Dependent variable
`X`	array-like (N*T, k)	Independent variables
`panel_id`	array-like (N*T,)	Panel unit identifiers
`time_id`	array-like (N*T,)	Time period identifiers
`var_names`	list of str	Regressor names
`dep_var`	str	Dependent variable name

Returns: CupFMResults

`model.summary()`¶

Print the summary table from the last fit.

`model.plot(kind='all', save_prefix=None)`¶

Generate plots. kind options: 'all', 'coefficients', 'factors', 'loadings', 'convergence', 'omega', 'ic'.

Results Class¶

`CupFMResults`¶

Returned by model.fit().

Attributes¶

Attribute	Type	Description
`beta`	ndarray (k,)	CupFM coefficients (primary)
`tstat`	ndarray (k,)	CupFM t-statistics
`pvalue`	ndarray (k,)	CupFM p-values
`betas`	dict	All 5 estimator coefficients
`tstats`	dict	All 5 t-statistics
`pvalues`	dict	All 5 p-values
`se`	dict	All 5 standard errors
`ci`	dict	All 5 confidence intervals
`F_hat`	ndarray (T, r)	Estimated factors
`Lambda`	ndarray (N, r)	Estimated loadings
`N`, `T`, `r`	int	Panel dimensions
`convergence_path`	ndarray	β across iterations

Methods¶

Method	Description
`summary()`	Print publication-quality table
`to_dataframe()`	Convert to pandas DataFrame
`to_latex(caption=None)`	Generate LaTeX table
`to_excel(filename)`	Export to Excel
`to_csv(filename)`	Export to CSV

Simulation¶

`simulate_panel(N, T, k, r, beta, ...)`¶

Simulate panel data from the BKN (2009) DGP.

from pycupfm import simulate_panel

sim = simulate_panel(N=20, T=40, k=1, r=2, beta=2.0, seed=42)

`monte_carlo(n_reps, N, T, ...)`¶

Run a full Monte Carlo experiment.

from pycupfm import monte_carlo

mc = monte_carlo(n_reps=100, N=20, T=40, verbose=True)

Data¶

`load_grunfeld()`¶

Load the classic Grunfeld (1958) investment dataset.

from pycupfm.datasets import load_grunfeld
df = load_grunfeld()  # N=10, T=20

Plotting Functions¶

All functions return matplotlib.figure.Figure.

Function	Description
`plot_coefficients(results)`	Forest plot with 95% CIs
`plot_factors(results)`	Common factor time series
`plot_loadings(results)`	Loading bar/scatter
`plot_convergence(results)`	Iteration path
`plot_omega_heatmap(results)`	Ω covariance heatmap
`plot_factor_ic(ic_values)`	Bai-Ng IC vs r
`plot_loadings_heatmap(results)`	N×r loading heatmap
`plot_all(results)`	All plots at once

Kernel Functions¶

from pycupfm import long_run_covariance, auto_bandwidth

Omega, Delta_plus = long_run_covariance(U, bandwidth=5, kernel='bartlett')
bw = auto_bandwidth(residuals, method='nw')

Available kernels: 'bartlett', 'parzen', 'qs' / 'quadratic_spectral'

API Reference¶

Main Class¶

CupFM¶

model.fit(y, X, panel_id, time_id, var_names=None, dep_var='y')¶

model.summary()¶

model.plot(kind='all', save_prefix=None)¶