Fisher Transform

I. What is the Fisher Transform Indicator

The Fisher Transform is a mathematical transformation indicator that converts price distributions into an approximately normal (Gaussian) distribution. Through a nonlinear transformation, it maps prices into an unbounded space, greatly stretching the distribution tails and producing sharper turning signals. This makes price reversal points much easier to identify.

Historical Background

The Fisher Transform was invented by John Ehlers and published in his 2002 book Cybernetic Analysis for Stocks and Futures as well as in Technical Analysis of Stocks and Commodities magazine. Ehlers is a pioneer in applying signal processing techniques to financial analysis, bringing Digital Signal Processing (DSP) concepts into technical analysis. The indicator is named after statistician Ronald Fisher, who developed the Fisher z-transformation.

Indicator Classification

• Type: Oscillator, displayed in a separate panel
• Category: Momentum / Statistical Transformation
• Default Parameter: Period $n = 9$
• Value Range: Unbounded, but typically stays between -3 and +3

---

II. Mathematical Principles and Calculation

Core Formula

Step 1: Normalize price to [-1, 1] range

$x_t = 2 \times \frac{C_t - LL_n}{HH_n - LL_n} - 1$

Where:

• $C_t$ = current closing price (or median price $(H+L)/2$)
• $LL_n$ = lowest low over the past $n$ periods
• $HH_n$ = highest high over the past $n$ periods

Step 2: Clamp and smooth

Clamp $x_t$ to $[-0.999, +0.999]$ to prevent logarithm overflow, and apply EMA smoothing:

$Value_t = 0.5 \times (x_t + Value_{t-1})$

Take the clamped value: $V_t = \min(\max(Value_t, -0.999), +0.999)$

Step 3: Apply Fisher Transform

$Fisher_t = 0.5 \times \ln\left(\frac{1 + V_t}{1 - V_t}\right)$

This formula is the inverse hyperbolic tangent function: $Fisher = \text{atanh}(V)$.

Step 4: Signal line

$Signal_t = Fisher_{t-1}$

The previous period’s Fisher value serves as the signal line.

Mathematical Meaning

The Fisher Transform $f(x) = 0.5 \cdot \ln\frac{1+x}{1-x}$ has these properties:

• When $x$ is near 0: $f(x) \approx x$ (linear)
• When $x$ approaches ±1: $f(x) \to \pm\infty$ (extreme amplification)

This means when price is in the middle of the range, the Fisher value changes gradually; but when price approaches extreme positions, the Fisher value changes dramatically, producing sharp peaks and troughs.

Step-by-Step Calculation Logic

1. Find range extremes: Highest high and lowest low over the past $n$ periods
2. Normalize: Map price to [-1, 1]
3. Smooth: Use simple recursive smoothing ($\alpha = 0.5$)
4. Clamp: Restrict values to ±0.999
5. Fisher Transform: Apply $0.5 \ln\frac{1+x}{1-x}$
6. Signal line: 1-period lag of Fisher

---

III. Python Implementation

import numpy as np
import pandas as pd


def fisher_transform(high: pd.Series,
                     low: pd.Series,
                     period: int = 9) -> pd.DataFrame:
    """
    Calculate Fisher Transform

    Parameters
    ----------
    high : pd.Series
        High price series
    low : pd.Series
        Low price series
    period : int
        Lookback period, default 9

    Returns
    -------
    pd.DataFrame
        DataFrame with Fisher and Signal columns
    """
    # Use median price
    median_price = (high + low) / 2

    highest = high.rolling(window=period, min_periods=period).max()
    lowest = low.rolling(window=period, min_periods=period).min()

    # Normalize to [-1, 1]
    raw = 2 * (median_price - lowest) / (highest - lowest) - 1

    # Smooth and clamp
    value = pd.Series(np.nan, index=high.index)
    fisher = pd.Series(np.nan, index=high.index)

    first_valid = raw.first_valid_index()
    if first_valid is None:
        return pd.DataFrame({"Fisher": fisher, "Signal": fisher}, index=high.index)

    start_idx = high.index.get_loc(first_valid)
    value.iloc[start_idx] = 0.0
    fisher.iloc[start_idx] = 0.0

    for i in range(start_idx, len(high)):
        if np.isnan(raw.iloc[i]):
            continue

        # Smooth
        prev_val = value.iloc[i - 1] if i > start_idx and not np.isnan(value.iloc[i - 1]) else 0.0
        v = 0.5 * raw.iloc[i] + 0.5 * prev_val

        # Clamp
        v = max(min(v, 0.999), -0.999)
        value.iloc[i] = v

        # Fisher Transform
        fisher.iloc[i] = 0.5 * np.log((1 + v) / (1 - v))

    signal = fisher.shift(1)

    return pd.DataFrame({
        "Fisher": fisher,
        "Signal": signal
    }, index=high.index)


# ========== Usage Example ==========
if __name__ == "__main__":
    np.random.seed(42)
    dates = pd.date_range("2024-01-01", periods=120, freq="D")
    base = 100 + np.cumsum(np.random.randn(120) * 1.0)

    df = pd.DataFrame({
        "date": dates,
        "open":  base + np.random.randn(120) * 0.3,
        "high":  base + np.abs(np.random.randn(120) * 0.8),
        "low":   base - np.abs(np.random.randn(120) * 0.8),
        "close": base,
        "volume": np.random.randint(1000, 10000, size=120),
    })
    df.set_index("date", inplace=True)

    # Calculate Fisher Transform
    ft = fisher_transform(df["high"], df["low"], period=9)
    df = pd.concat([df, ft], axis=1)

    print("=== Fisher Transform Results (Last 15 Rows) ===")
    print(df[["close", "Fisher", "Signal"]].tail(15).to_string())

    # Crossover signals
    df["buy"] = (df["Fisher"] > df["Signal"]) & (df["Fisher"].shift(1) <= df["Signal"].shift(1))
    df["sell"] = (df["Fisher"] < df["Signal"]) & (df["Fisher"].shift(1) >= df["Signal"].shift(1))

    print("\n=== Buy Signals (Fisher Crosses Above Signal) ===")
    print(df[df["buy"]][["close", "Fisher", "Signal"]].head(10).to_string())
    print("\n=== Sell Signals (Fisher Crosses Below Signal) ===")
    print(df[df["sell"]][["close", "Fisher", "Signal"]].head(10).to_string())

---

IV. Problems the Indicator Solves

1. Producing Sharp Turning Signals

Fisher Transform’s greatest advantage is making price reversal points very sharp and obvious. Traditional oscillators show relatively smooth changes in extreme areas, while Fisher Transform produces very sharp peaks and troughs that are easy to identify.

2. Crossover Signals

Signal	Condition	Meaning
Buy	Fisher crosses above Signal	Momentum shifts from bearish to bullish
Sell	Fisher crosses below Signal	Momentum shifts from bullish to bearish

3. Extreme Value Reversals

• Fisher > 1.5 ~ 2.0 -> Overbought zone; watch for Signal cross-under as sell signal
• Fisher < -1.5 ~ -2.0 -> Oversold zone; watch for Signal cross-over as buy signal

4. Approximately Normal Distribution

Since Fisher-transformed values are approximately normally distributed, standard deviation (e.g., ±2σ) can be used to define statistically-based overbought/oversold thresholds — more theoretically grounded than RSI’s empirical thresholds.

---

V. Advantages, Disadvantages, and Use Cases

Advantages

Advantage	Description
Sharp signals	Reversal points are very clear and hard to miss
Statistical foundation	Based on probability distribution transformation with mathematical theory support
Unbounded distribution	Not limited to a fixed range; extreme values are fully expressed
Fast response	Responds quickly to price changes

Disadvantages

Disadvantage	Description
Noise sensitive	High sensitivity brings more false signals
Few parameters	Only one period parameter, limited tuning space
Abstract concept	Fisher Transform’s mathematical meaning is less intuitive than RSI or MACD
Tail risk	Extreme values can be very large, affecting chart readability

Use Cases

• Short-term trading: Fisher’s sensitivity is well-suited for short-term and intraday trading
• Reversal trading: Capturing reversals within well-defined support/resistance ranges
• Pairing with other Ehlers indicators: Such as MAMA, Instantaneous Trendline, etc.

Comparison with Related Indicators

Comparison	Fisher	RSI	Stochastic
Distribution	Approx. normal	Skewed	Skewed
Value range	Unbounded	0–100	0–100
Signal sharpness	Very high	Medium	Medium
Response speed	Fast	Medium	Fast
False signal rate	High	Medium	High