Fisher Transform

A statistical transform that reshapes bounded price position into a sharper, near-Gaussian oscillator.

What it is

The Fisher Transform is a momentum oscillator introduced to trading by John Ehlers. It applies a genuine statistical function — the inverse hyperbolic tangent, 0.5 × ln((1 + x) / (1 − x)) — to a normalized measure of where price sits within its recent range. The idea is that raw price position is bounded and its distribution is compressed near the extremes; the Fisher Transform stretches those extremes out so that the resulting series looks closer to a normal (Gaussian) distribution, with sharp, well-defined turning points instead of the rounded tops and bottoms typical of smoothed oscillators. It is plotted in a separate pane, usually with a one-bar-delayed copy of itself as a trigger line. Important honesty note: the transform makes the indicator's values approximately normal, not the market — price returns famously have fat tails, and the normality assumption is weakest exactly when markets move violently. The Fisher Transform describes recent price position; it predicts nothing. Educational content only, not financial advice; trading involves risk of loss.

How it works

The calculation has three stages. First, normalize price position: take a source (Ehlers used the bar midpoint, hl2) and locate it within the highest and lowest source values over a lookback window (commonly 9-10 bars), rescaled to the range −1 to +1. Second, smooth that raw position with a light recursive filter (the classic coefficients are 0.66 on the new value and 0.33 on the previous smoothed value, often written 0.67 in code) and clamp it just inside ±1 — the clamp matters, because the transform goes to infinity at exactly ±1. Third, apply the Fisher Transform itself, 0.5 × ln((1 + x) / (1 − x)), and smooth the output with another light recursive average. The logarithmic stretch is what gives the indicator its character: small changes in position near the middle of the range barely move the output, while the same changes near the extremes move it a lot. That is why Fisher Transform peaks look sharp and 'spiky' compared with RSI or stochastics. The output is unbounded in principle, though in practice most readings fall roughly between −3 and +3, and the one-bar-lagged trigger line turns crossovers into discrete events.

How traders read it

• Extreme readings (for example beyond roughly ±1.5 to ±2) describe price sitting near the edge of its recent range after the transform's stretching — a description of a stretched condition, not a reversal signal.
• Crossovers between the Fisher line and its one-bar-delayed trigger are the conventional event people watch; they occur frequently, especially on short lookbacks, and many are noise.
• The zero line separates position in the upper versus lower half of the recent range; sustained readings on one side describe a directional phase.
• Sharp peaks: because of the logarithmic stretch, tops and bottoms in the indicator are more pointed than in bounded oscillators — this changes the shape, not the information content, which is still past price.
• Compare with the raw range position (a stochastic-style reading) to see exactly what the transform added: same underlying data, different distribution of values.

Common settings

Typical length is 9 or 10 bars applied to hl2 (the bar midpoint), with the trigger line as the Fisher value delayed one bar. Shorter lengths make the normalization window tighter and the output far more jumpy; longer lengths (20+) smooth the position estimate but delay everything. The smoothing coefficients (0.66/0.33-style) are Ehlers' conventions and are hard-coded in most implementations. As with any oscillator, changing length or source changes every reading, so state your settings when comparing results.

Strengths

• Grounded in a real, well-defined statistical transformation rather than ad-hoc arithmetic — every step can be verified.
• Produces sharp, clearly located turning points in the indicator line, which are easier to timestamp than the rounded extremes of RSI or stochastics.
• The near-Gaussian output makes thresholds (e.g. ±2) more comparable across instruments than raw range position would be.
• Cheap to compute and works on any market and timeframe.
• A good teaching tool: it makes the difference between transforming an indicator and changing the underlying information easy to see.

Pitfalls to watch

• The Gaussian claim applies to the indicator, not the market: real returns have fat tails, and the normality intuition breaks down precisely during the violent moves traders care most about.
• Sharp turns are not early turns: the transform reshapes the distribution of a lagging range-position measure; the input still only knows past prices.
• Trigger-line crossovers are frequent and whipsaw badly in ranges, particularly on the common 9-10 bar settings.
• Numerical edge cases: implementations must clamp the normalized value away from ±1 or the log term blows up — sloppy ports can produce spikes or NaNs.
• Settings sensitivity: the lookback window defines 'the range', and different lengths can put the same bar near an extreme or in the middle.
• No predictive power: an unusual Fisher reading is a description of recent price position, not a forecast, and no transform removes the risk of loss.

Pine v6 example

//@version=6
indicator("Fisher Transform Example", overlay = false)

length = input.int(9, "Length", minval = 1)
src    = input.source(hl2, "Source")

hi = ta.highest(src, length)
lo = ta.lowest(src, length)
rng = math.max(hi - lo, syminfo.mintick)

// Smoothed price position in [-1, 1], clamped away from the poles
value = 0.0
value := 0.66 * ((src - lo) / rng - 0.5) + 0.67 * nz(value[1])
value := math.min(math.max(value, -0.999), 0.999)

// Fisher Transform with light output smoothing
fish = 0.0
fish := 0.5 * math.log((1 + value) / (1 - value)) + 0.5 * nz(fish[1])

plot(fish, "Fisher", color = color.blue)
plot(fish[1], "Trigger", color = color.orange)
hline(0, "Zero", color = color.gray)

Pro tip: Plot the Fisher Transform next to a plain stochastic of the same length and source. You will see they turn at essentially the same bars — the transform changes the shape and scale of the extremes, not the timing of the information. That comparison keeps expectations honest: the statistics are real, the foresight is not. 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.