Driscoll-Kraay Standard Errors
Also known as: DK Standard Errors, Driscoll-Kraay Covariance Estimator, Spatial-Temporal HAC Standard Errors, Driscoll-Kraay Standart Hatalar
Driscoll-Kraay standard errors provide a nonparametric, heteroskedasticity- and autocorrelation-consistent (HAC) covariance estimator for balanced and unbalanced panel datasets. Introduced by Driscoll and Kraay in 1998, the method corrects inference when residuals exhibit cross-sectional dependence, serial autocorrelation, and heteroskedasticity simultaneously—problems common in macroeconomic and international finance panels where units such as countries or industries share common shocks.
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
Apply Driscoll-Kraay standard errors when working with macro panels—typically country-level, industry-level, or regional data—where the time dimension T is moderately large (at least 10–15 periods) and cross-sectional dependence is suspected. The estimator is appropriate alongside pooled OLS or fixed-effects models and does not require knowing the exact dependence structure. It is less suitable when T is very small (asymptotics are in T, not N), and it does not correct coefficient estimates—only inference. When N is large and T is short, two-way cluster-robust standard errors or a factor-model residual correction may be preferable.
Strengths & limitations
- Simultaneously corrects for cross-sectional dependence, heteroskedasticity, and serial autocorrelation without specifying the dependence structure.
- Consistent as T grows regardless of the size of N, making it well-suited for macro panels with many units.
- Compatible with any linear estimator (OLS, fixed effects, first differences) as a post-estimation sandwich correction.
- Widely implemented in Stata (xtscc), R (sandwich, plm), and Python (linearmodels), enabling straightforward application.
- Consistency relies on T → ∞; with small T the estimator can be unreliable and may undercover.
- Does not address endogeneity or omitted variables—only corrects the covariance matrix of an otherwise valid estimator.
- Bandwidth selection m(T) is rule-of-thumb (often floor(T^(1/4))); results can be sensitive to this choice.
- Not valid for nonlinear models or limited-dependent-variable panels without additional derivation.
Frequently asked
How does Driscoll-Kraay differ from two-way cluster-robust standard errors?
Two-way clustering adjusts for dependence within clusters defined by unit and time separately, relying on both N and T being large for asymptotics. Driscoll-Kraay aggregates scores across all units and applies a HAC kernel, so its asymptotics depend only on T → ∞ and it handles spatial dependence of unknown form without requiring discrete cluster groupings.
Can I use Driscoll-Kraay with random effects or instrumental variables?
The original derivation covers OLS and within-group (fixed-effects) estimators. Extensions to IV and GLS contexts are feasible by replacing the OLS residuals and score matrices with their IV or GLS analogues, but these require careful implementation and are not always available in standard software packages.
What bandwidth should I choose for the lag truncation m(T)?
The default rule of thumb is m(T) = floor(T^(1/4)), which Driscoll and Kraay (1998) recommend as a practical starting point. Researchers should also inspect residual autocorrelation functions and conduct sensitivity checks across a small range of bandwidth values to confirm that results are not highly sensitive to this choice.
Sources
- Driscoll, J. C., & Kraay, A. C. (1998). Consistent covariance matrix estimation with spatially dependent panel data. Review of Economics and Statistics, 80(4), 549–560. DOI: 10.1162/003465398557825 ↗
How to cite this page
ScholarGate. (2026, June 2). Driscoll-Kraay Standard Errors. ScholarGate. https://scholargate.app/en/econometrics/driscoll-kraay-se
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.
- Newey-West HACEconometrics↔ compare
- Pesaran CD TestEconometrics↔ compare