Pairs Trading Spread & Z-Score

A statistical construction that measures the gap between two related instruments and expresses it in standard deviations.

What it is

A pairs-trading spread is not a single built-in indicator but a statistical construction: you take two instruments that have historically moved together, estimate a hedge ratio that scales one against the other, and plot the difference between them. The idea, popularized by statistical-arbitrage desks in the 1980s, is that the spread between two related instruments may be more stable and more mean-reverting than either price on its own. The z-score is the second half of the construction: it normalizes the spread by its own rolling mean and standard deviation, so a reading of +2 means the spread is two standard deviations above its recent average. To build one in Pine Script you need three ingredients: a second symbol fetched safely with request.security, a rolling hedge ratio (a least-squares slope, computable from correlation and standard deviations), and a rolling z-score of the resulting spread. All of this is descriptive statistics about the recent past. A spread that has mean-reverted historically carries no guarantee it will continue to — relationships between instruments break, sometimes abruptly. This page is educational only, not financial advice, and any trading based on spreads carries real risk of loss.

How it works

Start with two price series: the chart symbol (call it Y) and a second symbol (X) retrieved with request.security. To avoid repainting, the second series should be requested with a one-bar offset and lookahead disabled, which means the spread is built from the other symbol's last completed bar. Next, estimate the hedge ratio (beta) — how many units of X offset one unit of Y. A rolling ordinary-least-squares slope can be computed directly on the chart as correlation(Y, X) × stdev(Y) / stdev(X) over a chosen lookback. The spread is then Y − beta × X. Finally, the z-score standardizes the spread: subtract its rolling mean and divide by its rolling standard deviation over a second lookback window. The result oscillates around zero, with ±2 commonly drawn as reference lines. Every stage involves a lookback choice, and each choice changes the output: a short hedge-ratio window makes beta jumpy; a long one makes it stale when the relationship shifts. Critically, the whole construction assumes the two instruments are cointegrated — that the spread is genuinely stationary. That property is estimated in-sample and routinely breaks out-of-sample, which is the honest core caveat of all pairs analysis: statistical correctness of the calculation says nothing about future stability of the relationship.

How traders read it

• Z-score near 0: the spread is close to its recent average — the two instruments are trading in line with their recent relationship.
• Z-score at +2 or −2: the spread is roughly two standard deviations from its rolling mean. Mean-reversion practitioners read this as 'stretched', but a stretched spread can keep stretching if the underlying relationship has changed.
• Direction of the spread: a rising spread means the chart symbol is strengthening relative to beta-times the second symbol; falling means the reverse. It says nothing about either instrument's absolute direction.
• Hedge-ratio stability: watch how much beta itself moves over time. A beta that drifts steadily or jumps suggests the historical relationship is not stable, which undermines any z-score reading built on it.
• Regime breaks: if the z-score sits beyond ±3 for an extended period rather than reverting, treat that as evidence the in-sample relationship may have broken, not as an ever-better opportunity.
• Always read the spread alongside what is happening in each instrument — earnings, rate decisions, or structural news can permanently re-price one leg.

Common settings

There are no universal defaults because this is a construction, not a packaged indicator. Common starting points are a hedge-ratio lookback of 100–250 bars and a z-score lookback of 20–100 bars, with reference lines at ±2. Some practitioners fix beta at 1 (a simple price difference or ratio) for closely related instruments; others re-estimate it on a rolling basis. Longer windows give steadier estimates that adapt slowly; shorter windows adapt fast but are noisy. Every choice should be tested for sensitivity rather than assumed.

Strengths

• Puts a fuzzy idea ('these two usually move together') into an exact, reproducible calculation that can be inspected line by line.
• The z-score gives a scale-free reading, so the same ±2 reference means the same thing across different pairs and price levels.
• Genuinely market-neutral in construction: the spread isolates relative movement rather than outright direction.
• Good statistics education: it forces explicit thinking about hedge ratios, stationarity, and lookback choices that most chart indicators hide.
• Straightforward to implement without repainting once request.security is handled correctly.

Pitfalls to watch

• Cointegration measured in-sample routinely breaks out-of-sample: a spread that mean-reverted for two years can trend for months when the underlying relationship changes. Correctness of the math does not imply profitability.
• Settings sensitivity: hedge-ratio and z-score lookbacks materially change every reading; a pair that looks 'stretched' on one window looks normal on another.
• Repaint risk in careless implementations: requesting the second symbol without a bar offset and with lookahead enabled produces z-scores that could never have been computed live.
• The z-score assumes the spread's distribution is roughly stable and normal-ish; financial spreads have fat tails, so ±2 events happen more often than a normal distribution implies.
• Stale beta: a hedge ratio estimated over old data silently misweights the legs, making the 'spread' partly a directional bet without the user realizing.
• No predictive power: the construction describes the recent statistical relationship only, and both legs carry full market risk if the relationship fails.

Pine v6 example

//@version=6
indicator("Pairs Spread Z-Score", overlay = false)

sym      = input.symbol("OANDA:EURGBP", "Second Symbol")
betaLen  = input.int(100, "Hedge Ratio Lookback", minval = 20)
zLen     = input.int(50, "Z-Score Lookback", minval = 10)

// Non-repainting fetch: prior completed bar, lookahead off
x = request.security(sym, timeframe.period, close[1], lookahead = barmerge.lookahead_off)
y = close

// Rolling OLS hedge ratio: beta = corr(y,x) * stdev(y) / stdev(x)
beta = ta.correlation(y, x, betaLen) * ta.stdev(y, betaLen) / ta.stdev(x, betaLen)

spread = y - beta * x
meanS  = ta.sma(spread, zLen)
stdS   = ta.stdev(spread, zLen)
z      = stdS != 0 ? (spread - meanS) / stdS : 0.0

plot(z, "Spread Z-Score", color = color.blue)
hline(2,  "+2", color = color.gray)
hline(0,  "Zero", color = color.new(color.gray, 50))
hline(-2, "-2", color = color.gray)

Pro tip: Before reading any z-score, plot beta itself and check whether it is stable over your sample. If the hedge ratio drifts materially, the spread is not measuring what you think it is, and the z-score inherits that error. Re-test the construction across different lookbacks and time periods rather than trusting one clean-looking chart — an in-sample relationship is a description of the past, not a forecast. This is educational material only, not financial advice; pairs relationships can and do break, and trading involves risk of loss.

Educational only — not financial advice, not a recommendation to trade. No indicator is predictive; trading involves substantial risk of loss.