Skip to main content

Quant Analytics: from scratch to pro · Factors, volatility, regimes and lead-lag · Pro

Cross-asset lead-lag, cointegration and spreads

Some markets seem to lead others, and some pairs of prices stay tied together. Both ideas are easy to see and easy to fake: non-synchronous closes create apparent lead-lag, and trending prices create apparent cointegration. This lesson shows how to measure each, and what a spread's half-life says.

4 min read Reviewed Revision 1

In this lesson

  • Measure lead-lag with cross-correlation and judge whether it is statistically real
  • Explain how non-synchronous closes and time zones create false lead-lag
  • Describe the Engle-Granger two-step test and what cointegration means
  • Compute the half-life and the standard deviation of a mean-reverting spread

Read first: Time series basics: random walks, autocorrelation and stationarity, Correlation and its limits, Volatility: from daily moves to an annual figure

Expected gap remaining as a spread reverts

  • AR(1) coefficient 0.95
  • AR(1) coefficient 0.98
0% 1% 2% 3% 05101520304060 Days after a two-standard-deviation gap
Expected gap. Computed from the expected path of an AR(1) spread starting at a 3.2% gap. The higher coefficient means a longer half-life, so capital waits longer for the gap to close.

Worked numbers

Each result is recomputed from the formula shown.

WhatFormulaResult
standard error of a correlation, 250 observations1 / sqrt(250)0.063
standard error, 1,000 observations1 / sqrt(1000)0.032
t-statistic of a lag correlation of 0.150.15 / se10004.74
variance explained0.15**22.25%
observations to detect a correlation of 0.10(2 / 0.10)**2400
hedge ratio from correlation 0.9 and volatilities 1.2% and 1.0%0.9 * 1.2 / 1.01.08
half-life of the spread, daysln(0.5) / ln(0.95)13.5
half-life at coefficient 0.98, daysln(0.5) / ln(0.98)34.3
stationary standard deviation of the spread0.005 / sqrt(1 - 0.95**2)1.60%
a two-standard-deviation gap2 * sd_spread3.20%
gap after round-trip costs of 0.30%gap2 - 0.0032.9%

Check yourself

The correlation between yesterday's return in market A and today's return in market B is 0.15 over 1,000 days. Is it distinguishable from zero?

The standard error is about 1 / sqrt(1000) = 0.032, so t is about 4.7. Statistically yes. It explains only 0.15^2 = 2.25% of the variance, and it may reflect different closing times, not a real lead.

Why can a market that closes earlier appear to lead one that closes later?

News arriving after the first close shows up in the first market's next return and in the second market's current return. A lag of one bar then looks like leadership, though both reacted to the same news.

Two prices each trend upward. Their regression has a high R-squared. Are they cointegrated?

Not necessarily. Two independent trending series can show a high R-squared (spurious regression). Cointegration needs the residual spread to be stationary, which is tested with unit-root tests.

A spread follows an AR(1) with coefficient 0.95 a day. What is its half-life?

ln(0.5) / ln(0.95) = 13.5 days.

Why can a pair that was cointegrated in the past stop being so?

The tie is a statistical property of a sample. Fundamentals change, and the relationship can break; the estimated hedge ratio also drifts.

Key terms

CointegrationLead-lagHalf-lifeAutocorrelation

Checked facts

Every figure below carries how it was checked: Primary checked (read in the cited primary, official, academic or data source), Two sources agree, or Approximate (an order of magnitude only).

  • Primary checked

    Gatev, Goetzmann and Rouwenhorst (2006) tested pairs trading on daily US stock data from 1962 to 2002 and reported average annualised excess returns of up to 11% for self-financing portfolios of pairs, above conservative transaction cost estimates.

    As of 2006. Source: Pairs Trading: Performance of a Relative-Value Arbitrage Rule.

Sources

  • Co-Integration and Error Correction: Representation, Estimation, and Testing. Econometrica 55(2), 251-276 (Robert F. Engle, C. W. J. Granger), 1987. Accessed 9 Nov 2026
  • Pairs Trading: Performance of a Relative-Value Arbitrage Rule. The Review of Financial Studies 19(3), 797-827 (Evan Gatev, William N. Goetzmann, K. Geert Rouwenhorst), 2006. Accessed 9 Nov 2026
  • Spurious Regressions in Econometrics. Journal of Econometrics 2(2), 111-120 (C. W. J. Granger, P. Newbold), 1974. Accessed 9 Nov 2026
  • Distribution of the Estimators for Autoregressive Time Series with a Unit Root. Journal of the American Statistical Association 74(366), 427-431 (David A. Dickey, Wayne A. Fuller), 1979. Accessed 9 Nov 2026

Informational content. Not investment advice.