Z-Score

A rolling statistical measure of how many standard deviations price sits from its recent mean.

What it is

A z-score is a standard tool from statistics applied to a price chart: it expresses how far the current value of a series is from its rolling mean, measured in units of the rolling standard deviation. A z-score of +2 means price is two standard deviations above its recent average; 0 means price sits exactly at the average; −1.5 means one and a half standard deviations below it. On a chart it is computed over a fixed lookback window — for example the last 50 bars — so both the mean and the standard deviation update as each bar closes, and the z-score is plotted as an oscillator around zero. Traders use it to normalize 'how stretched is price right now?' into a scale that is comparable across instruments and timeframes, and it is the same construction that underlies Bollinger %B and pairs-trading spreads. A z-score is purely descriptive: it restates recent price behavior in standardized units and says nothing about what happens next. Interpreting z-score extremes as automatic reversal points imports an assumption (mean reversion) that the number itself does not contain. This is educational content, not financial advice, and all trading involves risk of loss.

How it works

The calculation has three steps, all over the same lookback window of n bars. First, compute the rolling mean of the source series (usually the close): the simple average of the last n values. Second, compute the rolling standard deviation of those same n values — the square root of the average squared deviation from the mean, which measures how dispersed the recent data is. Third, standardize: z = (current value − rolling mean) / rolling standard deviation. The division is what makes the output comparable across markets: a $5 move means something very different on a $50 stock than on a $5,000 index, but 'two standard deviations of its own recent behavior' is a like-for-like statement. The textbook interpretation — about 95% of observations within ±2, about 99.7% within ±3 — comes from the normal distribution, and this is where honesty matters: financial returns are not normally distributed. They exhibit fat tails and volatility clustering, so |z| > 3 events occur far more often than a normal table predicts, and precisely during the volatile episodes when the reading would matter most. The z-score arithmetic is always correct; the Gaussian probability language often attached to it is not. Also note both the mean and the standard deviation are themselves lagging, window-dependent estimates: change the lookback and every reading changes.

How traders read it

• Magnitude: |z| near 0 means price is close to its recent average; |z| above 2 means price is stretched relative to its own recent dispersion. 'Stretched' is a description, not a signal to fade the move.
• Sign: positive z means price is above its rolling mean, negative means below — a normalized version of the same information a moving-average comparison gives.
• Cross-market comparison: because the output is in standard-deviation units, z-scores let you compare how extended two different instruments are on a common scale, which raw prices cannot do.
• Trend caution: in a sustained trend, price can hold a high positive or negative z-score for a long time, because the rolling mean keeps chasing price. A persistent extreme often describes a strong trend, not an imminent reversal.
• Distribution caution: do not translate z-values into probabilities using the normal curve — returns have fat tails, so ±3 readings are much more common than the Gaussian 0.3% would suggest.
• Related tools: Bollinger %B is a rescaled cousin of the same idea, and pairs-trading spreads apply the z-score to the difference between two instruments rather than to one price.

Common settings

A lookback of 20 to 100 bars on the closing price is typical; 20 matches the standard Bollinger Band window, while 50 or 100 gives a slower, smoother reading. Reference lines are usually drawn at ±1, ±2, and sometimes ±3. Shorter windows make both the mean and the standard deviation estimates noisier, so the z-score whips around more; longer windows are steadier but slower to reflect regime changes. There is no correct setting — the window defines what 'recent' means, and every choice is a trade-off you should test rather than assume.

Strengths

• Puts price displacement on a standardized, unitless scale that is comparable across instruments and timeframes.
• Transparent, textbook math — every component (mean, standard deviation, ratio) is auditable and deterministic.
• Generalizes cleanly: the same construction works on price, spreads, indicator values, or volume.
• Directly connected to widely used tools (Bollinger Bands, %B, pairs-trading spreads), so it deepens understanding of those as well.
• Genuine statistics education value: it makes the concepts of dispersion and normalization concrete on a chart.

Pitfalls to watch

• The normality assumption fails: financial returns have fat tails and clustered volatility, so Gaussian probability statements ('a 3-sigma event should be rare') materially understate how often extremes occur.
• Extreme readings are not reversal signals: in trends, price can sit beyond ±2 for many consecutive bars while the mean catches up — fading every extreme means fighting strong moves.
• Window sensitivity: the z-score depends entirely on the lookback; the same bar can read +1 on a 20-bar window and +2.5 on a 100-bar window. Neither is more 'true'.
• Both inputs lag: the rolling mean and standard deviation are backward-looking estimates, so the z-score describes the recent past, and after volatility shifts the denominator can be stale.
• Standard deviation is distorted by the very outliers it measures: one large bar inflates the denominator and mechanically shrinks subsequent z-scores, muting readings just after big moves.
• No predictive power: a z-score restates where price is relative to its own recent history. It assigns no probability to future direction, and no indicator removes the risk of loss.

Pine v6 example

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

length = input.int(50, "Lookback", minval = 2)
src    = input.source(close, "Source")

mean = ta.sma(src, length)
sd   = ta.stdev(src, length)
z    = sd != 0 ? (src - mean) / sd : na

plot(z, "Z-Score", color = color.teal)
hline(2,  "+2 SD", color = color.gray)
hline(0,  "Mean",  color = color.new(color.gray, 50))
hline(-2, "-2 SD", color = color.gray)

// Descriptive marker only — gated to confirmed bars so it cannot repaint
stretched = math.abs(z) > 2 and barstate.isconfirmed
plotshape(stretched, "Beyond 2 SD", style = shape.circle, location = location.top, color = color.orange, size = size.tiny)

Pro tip: Use the z-score as a normalizer, not an oracle: it is excellent for asking 'how unusual is this move relative to this instrument's own recent behavior?' and poor for predicting what follows. Before acting on any threshold, check how often your instrument historically stayed beyond it and for how long — on trending markets the answer is usually 'often, and for a while'. Keep the Gaussian intuition out of position sizing, because fat tails break it exactly when it matters. Educational only, not financial advice; all trading carries risk of loss.

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