Autocorrelation: Structure and Sources

College Depth 75 in the knowledge graph I know this Set as goal
Unlocks 4 downstream topics
autocorrelation time-series diagnostics

Core Idea

Autocorrelation (serial correlation) occurs when errors are correlated over time: Cov(uₜ, uₛ) ≠ 0 for t ≠ s, often following an AR(1) structure. Sources include omitted variables, model misspecification, or true dynamics. Autocorrelation does not bias OLS but inflates standard errors, invalidating inference.

Explainer

You already know that OLS assumes the errors are uncorrelated with each other — that's one of the core OLS assumptions. In cross-sectional data, this is often plausible: the measurement error on one person's wage has nothing to do with another's. But in time-series data, this assumption is routinely violated. If GDP was above trend last quarter, it tends to be above trend this quarter too. That persistence in the outcome bleeds into the residuals if your model doesn't fully explain it, creating autocorrelation — each error is correlated with its own past.

The most common pattern is AR(1) autocorrelation, where the error today is a scaled version of yesterday's error plus a new shock: uₜ = ρuₜ₋₁ + εₜ. The parameter ρ (rho) measures how persistent the correlation is. If ρ = 0.8, today's error is strongly predicted by yesterday's. If ρ = 0, errors are independent and you're fine. When autocorrelation exists, OLS still finds the same coefficient estimates — it remains unbiased — but the formula it uses to compute standard errors assumes independent errors, so those standard errors are wrong. Typically they are too small, making t-statistics too large and inference too confident.

The sources of autocorrelation give you a diagnostic roadmap. Omitted variables that are themselves persistent will inject their dynamics into your residuals — if you're modeling consumption but omit consumer sentiment (which drifts slowly), the omitted variable's autocorrelation becomes your residuals' autocorrelation. Model misspecification — for instance, fitting a linear trend to an exponentially growing series — leaves a systematic curved pattern in residuals, which appears as autocorrelation even if the underlying errors aren't. True dynamics are a third source: if the true model should include lagged Y on the right-hand side (because yesterday's outcome causes today's), omitting those lags forces the dynamic into the error term.

Understanding the lag structure matters because not all autocorrelation is AR(1). MA(1) errors (where this period's error depends on last period's shock but not last period's error) have a different pattern: significant autocorrelation at lag 1 only. Seasonal data can show autocorrelation at lag 12 (monthly) or lag 4 (quarterly). The autocorrelation function (ACF) and partial autocorrelation function (PACF) plots reveal these patterns — a slow decay in the ACF is diagnostic of AR structure, while a sharp cutoff points to MA structure. Before applying any correction (GLS, Newey-West standard errors, adding lags), diagnose the pattern carefully: the right fix depends on the right diagnosis.

Practice Questions 5 questions

Prerequisite Chain

Counting to 10Counting to 20Understanding ZeroThe Number ZeroCounting to FiveOne-to-One CorrespondenceCombining Small Groups Within 5Addition Within 10Addition Within 20Two-Digit Addition Without RegroupingTwo-Digit Addition with RegroupingAddition Within 100Repeated Addition as MultiplicationMultiplication Facts Within 100Division as Equal SharingDivision as Grouping (Measurement Division)Division: Grouping (Repeated Subtraction) ModelDivision: Fair Sharing ModelDivision as Equal SharingDivision as GroupingBasic Division FactsDivision Facts Within 100Two-Digit by One-Digit DivisionDivision with RemaindersRemainders and Quotients in DivisionDivision Word ProblemsIntroduction to Long DivisionFactors and MultiplesPrime and Composite NumbersEquivalent FractionsRelating Fractions and DecimalsDecimal Place ValueReading and Writing DecimalsComparing and Ordering DecimalsAdding and Subtracting DecimalsMultiplying DecimalsDividing DecimalsDividing FractionsMixed Number ArithmeticOrder of OperationsInteger Order of OperationsVariable ExpressionsCombining Like TermsOne-Step EquationsTwo-Step EquationsSolving Multi-Step EquationsEquations with Variables on Both SidesAngle Pairs: Complementary, Supplementary, and VerticalParallel Lines and TransversalsCorresponding AnglesAlternate Interior AnglesTriangle Angle Sum TheoremExterior Angle TheoremTriangle Inequality TheoremSimilar Triangles: AA SimilaritySimilar Triangles: SSS and SAS SimilarityProportions in Similar TrianglesRight Triangle Trigonometry IntroductionTrigonometric Ratios ReviewRadian MeasureConverting Between Degrees and RadiansThe Unit CircleGraphing Sine and CosineGraphing Tangent and Reciprocal Trigonometric FunctionsDerivatives of Trigonometric FunctionsAntiderivativesIndefinite IntegralsBasic Integration RulesRiemann SumsDefinite Integral DefinitionProbability Density Functions and Continuous DistributionsCumulative Distribution FunctionsContinuous Random VariablesNormal DistributionClassical OLS Assumptions (Gauss-Markov)Autocorrelation: Structure and Sources

Longest path: 76 steps · 439 total prerequisite topics

Prerequisites (2)

Leads To (1)