Best Linear Unbiased Estimator (BLUE): The Gauss-Markov Theorem Explained

The "best linear unbiased estimator", usually shortened to BLUE, is one of the most important ideas in statistics and econometrics — it's the property that makes ordinary least squares (OLS) regression so widely trusted. Understanding BLUE means understanding why OLS is a good way to estimate relationships, and when it stops being so. This guide explains what BLUE means, the Gauss-Markov theorem behind it, the assumptions required, and what happens when they break — in clear, plain language. It's relevant to anyone studying econometrics, statistics or quantitative finance.

What does "best linear unbiased estimator" mean?

The phrase packs three separate properties into one acronym. An estimator is a rule for calculating an estimate of an unknown parameter (such as the slope in a regression) from data. For it to be BLUE, it must be:

• Linear — the estimator is a linear function of the observed data (the dependent variable).
• Unbiased — on average, across many samples, it gives the true value of the parameter; its expected value equals the parameter being estimated.
• Best — among all linear unbiased estimators, it has the smallest variance. "Best" here means most efficient — the least scattered, most precise estimate.

So "best linear unbiased estimator" is the linear, unbiased estimator that is as precise as possible — you can't do better in terms of variance without giving up linearity or unbiasedness.

Unbiased versus efficient: the intuition

It helps to separate two ideas that BLUE combines. Unbiasedness is about being right on average: if you repeated the study many times, the estimates would centre on the true value rather than systematically over- or under-shooting. Efficiency (the "best" part) is about being right consistently: the estimates don't scatter widely from sample to sample. You can have one without the other — an estimator can be unbiased but wildly variable, or very stable but systematically wrong. BLUE asks for both: centred on the truth and as tightly clustered as any linear unbiased rule can manage. That combination is exactly what makes an estimator trustworthy in practice.

The Gauss-Markov theorem

The reason BLUE matters is the Gauss-Markov theorem, one of the cornerstones of regression analysis. It states that, provided a set of assumptions holds, the ordinary least squares (OLS) estimator is BLUE — the best linear unbiased estimator of the regression coefficients. In other words, under those conditions, no other linear unbiased estimator can beat OLS for precision. This is a powerful justification for using OLS: it's not just convenient, it's optimal within its class when the assumptions are satisfied.

The Gauss-Markov assumptions

For OLS to be BLUE, the classical Gauss-Markov assumptions must hold:

• Linearity — the model is linear in its parameters.
• Random sampling — the data are drawn from the population appropriately.
• No perfect multicollinearity — the explanatory variables aren't perfectly correlated with each other.
• Exogeneity — the error term has a mean of zero given the explanatory variables (the regressors are uncorrelated with the errors).
• Homoskedasticity — the errors have constant variance.
• No autocorrelation — the errors aren't correlated with each other.

Notably, the Gauss-Markov theorem does not require the errors to be normally distributed. Normality is needed for certain hypothesis tests, but not for OLS to be BLUE.

What happens when the assumptions break

The interesting cases arise when assumptions fail. If there's heteroskedasticity (non-constant error variance) or autocorrelation (correlated errors), OLS generally remains unbiased — but it's no longer best. It loses its minimum-variance property, so other estimators (such as generalised least squares, GLS) can be more efficient, and the usual standard errors become unreliable. If exogeneity fails — for example because of omitted variables or simultaneity — OLS becomes biased altogether, which is more serious. Knowing which assumption has broken tells you whether you have an efficiency problem or a bias problem, and which fix to reach for.

Why BLUE matters

BLUE matters because it provides the theoretical backbone for the most widely-used estimation method in all of empirical economics and finance. It explains why we can rely on OLS when conditions are right, gives us a clear checklist of assumptions to test, and tells us what's at stake — bias or inefficiency — when each assumption is violated. That diagnostic clarity is what makes the concept so valuable in practice.

Frequently asked questions

What does BLUE stand for?

Best Linear Unbiased Estimator: a linear estimator that is unbiased (correct on average) and "best" in the sense of having the smallest variance among all linear unbiased estimators.

What is the Gauss-Markov theorem?

It states that, under the classical assumptions, the OLS estimator is the best linear unbiased estimator (BLUE) of the regression coefficients — no linear unbiased estimator is more precise.

Do the errors need to be normally distributed for OLS to be BLUE?

No. Normality is needed for some hypothesis tests, but not for the Gauss-Markov theorem; OLS is BLUE without assuming normal errors.

What happens if assumptions are violated?

Heteroskedasticity or autocorrelation leave OLS unbiased but no longer best (efficient). A failure of exogeneity makes OLS biased, which is more serious and requires different methods.

Master econometrics with Learnsignal

Concepts like BLUE underpin quantitative analysis. Learnsignal's tutor-led ACCA and CIMA courses build the analytical foundations — with flexible, supported online study that fits around work.