Serial Correlation Tests

What Is Serial Correlation?

Serial correlation (autocorrelation) occurs when the error terms of a panel model are correlated across time within the same entity:

\[\text{Cov}(\varepsilon_{it}, \varepsilon_{is}) \neq 0 \quad \text{for } t \neq s\]

In panel data, the most common form is first-order autocorrelation (AR(1)), where:

\[\varepsilon_{it} = \rho \, \varepsilon_{i,t-1} + v_{it}, \quad |\rho| < 1\]

Consequences of Ignoring Serial Correlation

• OLS coefficient estimates remain consistent but become inefficient
• Standard errors are biased (typically downward), leading to inflated t-statistics
• Confidence intervals are too narrow, causing over-rejection of null hypotheses
• R-squared and F-statistics become unreliable

Available Tests

PanelBox provides three complementary tests for detecting serial correlation:

Test	H₀	Order	Model Type	Best For
Wooldridge AR	No AR(1)	First-order	FE/RE	Default first test
Breusch-Godfrey	No AR(p)	Higher-order	Any	Detecting higher-order AR
Baltagi-Wu LBI	No AR(1)	First-order	Unbalanced	Unbalanced panels

Recommended Workflow

graph TD
    A[Estimate Panel Model] --> B[Run Wooldridge AR Test]
    B -->|Reject H₀| C[Serial Correlation Detected]
    B -->|Fail to Reject| D[No Evidence of AR1]
    C --> E[Run Breusch-Godfrey with Multiple Lags]
    E --> F{Higher-Order AR?}
    F -->|Yes| G[Model AR Structure or Use HAC SE]
    F -->|No, Just AR1| H[Use Clustered or Newey-West SE]
    D --> I[Proceed with Standard Inference]

Step-by-Step Testing

from panelbox import FixedEffects
from panelbox.datasets import load_grunfeld
from panelbox.validation.serial_correlation.wooldridge_ar import WooldridgeARTest
from panelbox.validation.serial_correlation.breusch_godfrey import BreuschGodfreyTest
from panelbox.validation.serial_correlation.baltagi_wu import BaltagiWuTest

# Load data and estimate model
data = load_grunfeld()
fe = FixedEffects(data, "invest", ["value", "capital"], "firm", "year")
results = fe.fit()

# Step 1: Wooldridge test (default first check)
wooldridge = WooldridgeARTest(results)
w_result = wooldridge.run(alpha=0.05)
print(f"Wooldridge F-stat: {w_result.statistic:.3f}, p-value: {w_result.pvalue:.4f}")
print(w_result.conclusion)

# Step 2: If rejected, check for higher-order AR
if w_result.reject_null:
    bg = BreuschGodfreyTest(results)
    for lag in [1, 2, 3]:
        bg_result = bg.run(lags=lag)
        print(f"  BG AR({lag}): LM={bg_result.statistic:.3f}, p={bg_result.pvalue:.4f}")

# Step 3: For unbalanced panels, also check Baltagi-Wu
bw = BaltagiWuTest(results)
bw_result = bw.run()
print(f"Baltagi-Wu z-stat: {bw_result.statistic:.3f}, p-value: {bw_result.pvalue:.4f}")
print(f"  Estimated rho: {bw_result.metadata['rho_estimate']:.4f}")

What to Do If Serial Correlation Is Detected

Option 1: Clustered Standard Errors (Recommended)

Clustering by entity accounts for arbitrary within-entity correlation:

results_cluster = fe.fit(cov_type="clustered")
print(results_cluster.summary())

Option 2: Newey-West (HAC) Standard Errors

Robust to both heteroskedasticity and autocorrelation:

results_nw = fe.fit(cov_type="newey_west")

Option 3: Driscoll-Kraay Standard Errors

Robust to serial correlation, heteroskedasticity, and cross-sectional dependence:

results_dk = fe.fit(cov_type="driscoll_kraay")

Option 4: Model the AR Structure

For dynamic panels, incorporate the autoregressive structure explicitly:

from panelbox.gmm import SystemGMM

# System GMM handles AR dynamics directly
gmm = SystemGMM(
    data, "invest", ["invest"], ["value", "capital"],
    "firm", "year", max_lags=2
)
gmm_results = gmm.fit()

Interpreting Results

All serial correlation tests return a ValidationTestResult with:

result.test_name       # Name of the test
result.statistic       # Test statistic (F, LM, or z)
result.pvalue          # p-value
result.df              # Degrees of freedom
result.reject_null     # True if H₀ rejected at chosen alpha
result.conclusion      # Human-readable conclusion
result.metadata        # Test-specific additional information

p-value	Decision	Action
< 0.01	Strong rejection	Use robust SE (clustered or HAC)
0.01 -- 0.05	Rejection	Use robust SE
0.05 -- 0.10	Borderline	Consider robust SE as a precaution
> 0.10	Fail to reject	Standard SE likely adequate

Software Equivalents

PanelBox	Stata	R (plm)
WooldridgeARTest	xtserial y x1 x2	plm::pwartest()
BreuschGodfreyTest	xttest1	plm::pbgtest()
BaltagiWuTest	xtserial (variant)	plm::pbltest()

See Also

• Heteroskedasticity Tests -- testing for non-constant variance
• Cross-Sectional Dependence Tests -- testing for correlation across entities
• Newey-West Standard Errors -- HAC standard errors
• Clustered Standard Errors -- cluster-robust inference

References

• Wooldridge, J. M. (2002). Econometric Analysis of Cross Section and Panel Data. MIT Press, Section 10.4.1.
• Breusch, T. S. (1978). "Testing for autocorrelation in dynamic linear models." Australian Economic Papers, 17(31), 334-355.
• Godfrey, L. G. (1978). "Testing against general autoregressive and moving average error models when the regressors include lagged dependent variables." Econometrica, 46(6), 1293-1301.
• Baltagi, B. H., & Wu, P. X. (1999). "Unequally spaced panel data regressions with AR(1) disturbances." Econometric Theory, 15(6), 814-823.
• Drukker, D. M. (2003). "Testing for serial correlation in linear panel-data models." Stata Journal, 3(2), 168-177.