{ "id": "0d0d7826-d6b2-56f2-b366-4b458d333268", "slug": "autocorrelation", "term": "Autocorrelation", "aliases": [], "category": "Quantitative Finance", "category_slug": "quantitative-finance", "difficulty": "intermediate", "definition": "Autocorrelation (also called serial correlation) is a statistical measure of the correlation between a time series and a lagged version of itself, quantifying the degree to which current values of a variable are linearly related to past values. In financial markets, autocorrelation in returns challenges the efficient market hypothesis, underpins momentum and mean-reversion strategies, and is a critical diagnostic tool for assessing the quality of quantitative trading models.", "key_takeaways": [ "Autocorrelation coefficient ρ(k) ranges from -1 to +1; positive autocorrelation indicates return momentum (today's gain predicts tomorrow's gain), negative autocorrelation indicates mean reversion.", "The Durbin-Watson statistic (range 0–4) tests for first-order autocorrelation in regression residuals: values near 2 indicate no autocorrelation; values near 0 indicate positive autocorrelation; values near 4 indicate negative autocorrelation.", "Price autocorrelation in liquid equity markets is generally near zero at daily frequencies (consistent with EMH), but significant momentum autocorrelation exists at weekly-to-12-month horizons and mean-reversion at multi-year horizons.", "Autocorrelation in hedge fund returns is itself a red flag: smoothed or stale pricing of illiquid assets creates artificially positive serial correlation, understating true volatility and Sharpe ratios.", "The Ljung-Box Q-statistic formally tests the null hypothesis of no autocorrelation across multiple lags simultaneously, widely used in ARIMA model diagnostics." ], "detailed_explanation": "Autocorrelation is defined as the correlation of a time series {r_t} with its own past values. The autocorrelation function (ACF) at lag k is: ρ(k) = Cov(r_t, r_{t-k}) / Var(r_t). The partial autocorrelation function (PACF) at lag k measures the correlation between r_t and r_{t-k} after removing the influence of intermediate lags. Together, the ACF and PACF are the primary diagnostic tools for identifying the structure of time series processes—essential in ARIMA modeling.\n\nIn financial returns, the presence and sign of autocorrelation has profound implications for investment strategy. Positive autocorrelation at short lags suggests momentum: assets that have recently risen tend to continue rising. This is the empirical foundation of cross-sectional and time-series momentum strategies, which have been documented across asset classes. Negative autocorrelation at longer lags suggests mean reversion: assets that have significantly outperformed tend to subsequently underperform, consistent with the DeBondt-Thaler reversal anomaly. These two phenomena can coexist because they operate at different time scales—momentum dominates at 1-12 month horizons, while mean reversion tends to emerge over 3-5 year windows.\n\nFor model diagnostics, autocorrelation in regression residuals is a serious problem. When a pricing or factor model generates autocorrelated residuals, it indicates that the model is systematically missing a pattern that could be captured—the residuals contain predictive information, meaning the model is misspecified. This violates the classical OLS assumption of serially uncorrelated errors and causes the reported standard errors to be incorrect (typically understated), leading to inflated t-statistics and false confidence in the model's factors.\n\nAutocorrelation in hedge fund return series is a particularly important diagnostic for due diligence. Asness, Krail, and Liew (2001) demonstrated that many hedge funds report artificially smooth returns due to stale pricing of illiquid assets—this creates high positive autocorrelation in reported monthly returns that has nothing to do with actual trading skill. Adjusting for this smoothing (using the Geltner unsmoothing technique) typically increases estimated volatility significantly and reduces the Sharpe ratio. Institutional allocators now routinely test for this by regressing current fund returns on lagged market returns; significant exposure to lagged market movements implies stale pricing.", "example": "A quantitative analyst examines monthly returns of a credit-focused hedge fund and finds first-order autocorrelation of 0.48 and second-order autocorrelation of 0.31, with the Ljung-Box Q(12) statistic strongly rejecting the null of no autocorrelation (p < 0.001). Applying the Geltner unsmoothing formula: σ_true = σ_reported / √(1 - 2ρ₁ + ρ₁²) ≈ σ_reported × 1.52. The fund's reported annualized volatility of 5.2% implies true volatility of approximately 7.9%. The reported Sharpe ratio of 1.45 falls to approximately 0.95 after adjustment—still acceptable but meaningfully lower, and the fund's apparent low correlation to equities is partially an artifact of return smoothing. The analyst recommends a capital allocation 30% smaller than the reported Sharpe ratio would suggest, accounting for the liquidity risk embedded in the return smoothing.", "formula": "ACF: ρ(k) = Cov(r_t, r_{t-k}) / Var(r_t)\nDurbin-Watson: DW = Σ(eₜ - eₜ₋₁)² / Σeₜ²\nGeltner Unsmoothing: r_true(t) = (r_reported(t) - α × r_reported(t-1)) / (1 - α), where α is first-order autocorrelation", "formula_latex": null, "interactive_type": "chart", "calculator_id": null, "related_terms": [ "correlation", "efficient-market-hypothesis", "factor-model", "factor-signal", "hedge-fund", "hurst-exponent", "liquidity", "liquidity-risk", "mean-reversion", "quantitative-analysis", "random-walk", "reversal", "serial-correlation", "sharpe-ratio", "support-vector-machine" ], "backlinks": [ "arima-model", "central-limit-theorem", "geometric-brownian-motion", "hurst-exponent", "ordinary-least-squares", "quantitative-analysis", "random-forest", "random-walk", "regression-analysis", "serial-correlation", "sharpe-ratio-annualized", "statistical-arbitrage", "time-series-analysis", "trend-following" ], "cross_references": [ "correlation", "efficient-market-hypothesis", "factor-model", "hedge-fund", "liquidity", "liquidity-risk", "mean-reversion", "reversal", "serial-correlation", "sharpe-ratio", "time-series-momentum", "volatility" ], "tags": [ "level:intermediate", "cat:quantitative-finance" ], "asset_classes": [], "regulators": [], "see_also": [], "sources": [], "wordcount": 705, "checksum": "efde1963d495fbf9", "version": "2026.05.03", "license": "CC-BY-4.0", "updated_at": "2026-09-07T02:15:24+00:00", "_links": { "self": "https://hedgefund.wiki/api/v1/terms/autocorrelation", "jsonld": "https://hedgefund.wiki/api/v1/terms/autocorrelation?format=jsonld", "markdown": "https://hedgefund.wiki/api/v1/terms/autocorrelation?format=md", "graph": "https://hedgefund.wiki/api/v1/graph/autocorrelation", "category": "https://hedgefund.wiki/api/v1/categories/quantitative-finance", "schema": "https://hedgefund.wiki/schema/term.schema.json", "html": "https://hedgefund.wiki/#/terms/autocorrelation" } }