FE - Fixed Effects
Overview
The Fixed Effects (FE) estimator handles panel data models with entity-specific unobserved heterogeneity. It uses within-transformation (demeaning) to eliminate time-invariant fixed effects and provides consistent estimates when individual effects are correlated with regressors.
Mathematical Foundation
Panel Data Model
where \(\alpha_i\) are entity-specific fixed effects and \(i = 1, \ldots, N\), \(t = 1, \ldots, T\).
Within Transformation
Eliminate fixed effects by demeaning: \(\(\tilde{y}_{it} = y_{it} - \bar{y}_i, \quad \tilde{\mathbf{x}}_{it} = \mathbf{x}_{it} - \bar{\mathbf{x}}_i\)\)
where \(\bar{y}_i = \frac{1}{T_i}\sum_{t=1}^{T_i} y_{it}\) and \(\bar{\mathbf{x}}_i = \frac{1}{T_i}\sum_{t=1}^{T_i} \mathbf{x}_{it}\).
FE Estimator
Apply OLS to demeaned data: \(\(\hat{\mathbf{\beta}}_{FE} = (\tilde{\mathbf{X}}'\tilde{\mathbf{X}})^{-1}\tilde{\mathbf{X}}'\tilde{\mathbf{y}}\)\)
Fixed Effects Recovery
API Reference
Constructor
Parameters:
- robust (bool): Whether to use cluster-robust standard errors
Methods
fit()
Fit the Fixed Effects model using within transformation.Parameters:
- X (np.ndarray): Training data matrix
- y (np.ndarray): Target values
- entity_id (np.ndarray): Entity identifiers for grouping
Requirements: - At least 2 observations per entity for within variation - n_samples > n_features + n_entities
predict()
Generate predictions including entity fixed effects.Statistical Methods
standard_errors()→ Standard errors (clustered if robust=True)t_statistics()→ T-statistics for significance testsp_values()→ P-values for coefficient testsconfidence_intervals(alpha=0.05)→ Confidence intervalssummary()→ Comprehensive regression summary
Properties
Core Results:
- coefficients: Fixed Effects regression coefficients
- fixed_effects: Entity-specific fixed effects (α̂ᵢ)
- robust: Whether clustered standard errors are used
Statistical Measures:
- r_squared: Overall R-squared
- within_r_squared: Within R-squared
- r_squared_adj: Adjusted R-squared
- mse: Mean squared error
- residuals: Regression residuals
Model Information:
- n_samples: Number of observations
- n_features: Number of features
- n_entities: Number of entities
Implementation Details
Within Transformation
- Efficient entity mapping for demeaning
- Vectorized computation of entity means
- Memory-optimized transformation
Standard Errors
- Classical: Assumes independence across entities
- Clustered: Accounts for within-entity correlation
Usage Guidelines
Use FE when: - Panel data structure present - Entity-specific unobserved heterogeneity - Fixed effects may correlate with regressors - Focus on within-entity variation
Key Assumptions: - Strict exogeneity: \(E[\epsilon_{it}|\mathbf{x}_{i1}, \ldots, \mathbf{x}_{iT}, \alpha_i] = 0\) - Time-varying regressors available - Sufficient within-entity variation
Limitations: - Cannot estimate effects of time-invariant variables - Requires balanced or unbalanced panel structure - Need sufficient time periods per entity
Consider alternatives when: - Time-invariant regressors of interest → Random Effects - Cross-sectional data → OLS - Endogenous regressors → IV or TSLS