PANIC Test: Panel Unit Root Analysis with Common Factor Decomposition
PANIC: Panel Analysis of Non-stationarity in Idiosyncratic and Common Components · Also known as: Panel Analysis of Non-stationarity in Idiosyncratic and Common Components, Bai-Ng PANIC Test, Second-Generation Panel Unit Root Test, Panel Birim Kök Testi (PANIC)
PANIC (Panel Analysis of Non-stationarity in Idiosyncratic and Common Components) is a second-generation panel unit root test introduced by Bai and Ng (2004). It decomposes each panel series into common factors and idiosyncratic components, then tests for unit roots in each part separately, making it robust to cross-sectional dependence — a critical limitation of first-generation tests such as IPS or LLC.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use PANIC when you have a moderately large panel (N and T both at least 10–20) suspected of cross-sectional dependence driven by unobservable common shocks, such as global business cycles, oil price shocks, or monetary policy spillovers. It is most appropriate when you want to distinguish between common-factor non-stationarity and unit-specific non-stationarity. PANIC is not suited for small panels, strongly unbalanced data, or situations where the factor structure is implausible. As an alternative, consider CIPS (Pesaran 2007) for simpler implementation without explicit factor extraction, or CADF for single-equation cross-sectionally augmented approaches.
Strengths & limitations
- Explicitly models and removes cross-sectional dependence via approximate factor structure, correcting the size distortions that plague first-generation tests
- Tests stationarity of common factors and idiosyncratic components separately, yielding richer diagnostic information about the source of non-stationarity
- Pooled P_a and P_b statistics have standard limiting distributions (chi-squared and normal), facilitating straightforward critical-value tables
- Applicable under heterogeneous factor loadings and heterogeneous short-run dynamics across units
- Requires the number of common factors r to be pre-specified or consistently estimated (e.g., via Bai-Ng IC criteria), and results can be sensitive to this choice
- Both N and T must be sufficiently large for the principal-component estimator and the asymptotic theory to be reliable; small-panel performance is not guaranteed
- Assumes an approximate factor model; if cross-sectional dependence has a non-factor structure (e.g., spatial or network dependence), the test may not fully account for it
- Estimation of idiosyncratic components introduces generated-regressor bias in small samples, which can affect finite-sample size
Frequently asked
How does PANIC differ from the CIPS test of Pesaran (2007)?
Both are second-generation tests robust to cross-sectional dependence, but PANIC explicitly extracts and tests common factors via principal components, then tests idiosyncratic components separately. CIPS instead augments individual ADF regressions with cross-sectional means as proxies for one common factor, which is simpler but less flexible when multiple common factors are present. PANIC is preferred when the factor structure requires more than one common trend.
What happens if I use PANIC on a panel with only cross-sectional dependence of the spatial type?
PANIC is designed for factor-structured dependence. If dependence follows a spatial or network pattern rather than a low-rank factor structure, principal-component extraction will not fully purge the cross-sectional correlation. Residual dependence will inflate test size. In such cases, spatial panel unit root tests or CIPS with heteroskedasticity-robust critical values may be more appropriate.
Should I report P_a, P_b, or both?
Bai and Ng (2004) recommend reporting both, since they pool information differently: P_a weights each series equally in log-p space while P_b uses the inverse-normal transformation. If results diverge, it may indicate tail sensitivity or outlier p-values in individual ADF tests, and closer inspection of unit-specific results is warranted. Reporting both is good practice for transparency.
Sources
- Bai, J., & Ng, S. (2004). A PANIC attack on unit roots and cointegration. Econometrica, 72(4), 1127–1177. DOI: 10.1111/j.1468-0262.2004.00528.x ↗
How to cite this page
ScholarGate. (2026, June 2). PANIC: Panel Analysis of Non-stationarity in Idiosyncratic and Common Components. ScholarGate. https://scholargate.app/en/econometrics/panic-test
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- CADF TestEconometrics↔ compare
- CIPS TestEconometrics↔ compare
- Dynamic Factor ModelEconometrics↔ compare