GarmanKlassVolatility

OHLC realised-volatility estimator. ~7.4× more statistically efficient than close-to-close stddev under driftless GBM, but biased when overnight drift is significant.

Quick reference

Field	Value
Family	Volatility & Bands
Input type	Candle (uses open, high, low, close)
Output type	f64
Output range	[0, ∞) (annualised percent)
Default parameters	period = 20, trading_periods = 252
Warmup period	period (exact)
Interpretation	Annualised realised volatility blending range and body.

Formula

s_t   = 0.5 · (ln(H_t / L_t))² − (2·ln 2 − 1) · (ln(C_t / O_t))²
sigma = √max(mean(s_t over `period`), 0)
out   = sigma · √trading_periods · 100

Garman & Klass (1980) extended Parkinson's high-low estimator by adding an open-to-close term, partially correcting the bias Parkinson picks up when the closing price drifts within the bar. The open-to-close weight is the constant 2·ln 2 − 1 ≈ 0.3863 (GK_OC_COEFF in the source). Under driftless GBM the estimator is ~7.4× more statistically efficient than the close-to-close stddev (Parkinson sits at ~5.0×).

The per-bar sample s_t can be marginally negative on narrow-range bars with large open-to-close moves; this matches the original paper's algebra and is handled by the max(·, 0) clamp on the rolling mean.

Parameters

Name	Type	Default	Constraint	Source
period	usize	20	>= 1	GarmanKlassVolatility::new (garman_klass.rs:74)
trading_periods	usize	252	>= 1	garman_klass.rs:74

Either parameter == 0 returns [Error::PeriodZero]. Python defaults come from #[pyo3(signature = (period=20, trading_periods=252))]; the Node constructor takes both arguments explicitly.

Inputs / Outputs

use wickra::{Indicator, GarmanKlassVolatility, Candle};
// GarmanKlassVolatility: Input = Candle, Output = f64
const _: fn(&mut GarmanKlassVolatility, Candle) -> Option<f64> = <GarmanKlassVolatility as Indicator>::update;

• Python. update(candle) returns float | None; batch(open, high, low, close) returns a 1-D float64 np.ndarray with NaN warmup.
• Node. update(open, high, low, close) returns number | null; batch(open, high, low, close) returns an Array<number> with NaN warmup.

Warmup

warmup_period() returns period; the first output lands on bar period (index period − 1). Pinned by first_emission_at_warmup_period (period 5: bars 1–4 return None, bar 5 emits).

Edge cases

• Zero-movement bars (O == H == L == C). Both log terms are zero; the estimator returns 0 (test zero_movement_yields_zero).
• Constant bar shape. Identical OHLC ratios give a constant output (test constant_bar_shape_yields_constant_sigma).
• Drift bias. Unbiased only under driftless GBM. For data with significant overnight drift (gaps), use YangZhangVolatility instead.
• Reset. reset() clears the window, running sum and last value.

Examples

Rust

use wickra::{BatchExt, Candle, GarmanKlassVolatility, Indicator};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let candles: Vec<Candle> = (0..40)
        .map(|i| {
            let base = 100.0 + f64::from(i);
            Candle::new(base, base + 2.0, base - 2.0, base + 0.5, 1.0, i64::from(i)).unwrap()
        })
        .collect();
    let mut gk = GarmanKlassVolatility::new(20, 252)?;
    println!("{:?}", gk.batch(&candles).into_iter().flatten().last());
    Ok(())
}

Python

import numpy as np
import wickra as ta

gk = ta.GarmanKlassVolatility(20, 252)
out = gk.batch(open_, high, low, close)  # 1-D annualised-% series, NaN warmup

Node

const ta = require('wickra');
const gk = new ta.GarmanKlassVolatility(20, 252);
const v = gk.update(100, 102, 98, 100.5); // null during warmup, else annualised %

Interpretation

Garman-Klass squeezes more information out of each bar than a close-to-close stddev by combining the high-low range with the open-to-close body:

1. Efficiency. For the same statistical confidence it needs roughly a seventh of the samples a close-to-close estimator would — useful when you want a responsive volatility read from limited history.
2. When to prefer it. Best on near-continuous intraday data with little overnight drift. On trending or gappy data its bias shows; reach for RogersSatchellVolatility (drift) or YangZhangVolatility (gaps) instead.

Common pitfalls

• Applying it to trending markets. A persistent drift biases the open-to-close term; Garman-Klass is a driftless-GBM estimator.
• Reading the negative intermediate samples as a bug. Individual s_t can be slightly negative by design; only the clamped rolling mean is taken to the square root.

References

• Mark B. Garman & Michael J. Klass, On the Estimation of Security Price Volatilities from Historical Data, The Journal of Business, vol. 53, no. 1, 1980, pp. 67–78.

See also

• ParkinsonVolatility — high-low only.
• RogersSatchellVolatility — drift-free OHLC alternative.
• YangZhangVolatility — adds overnight variance; recommended for data with gaps.
• HistoricalVolatility — close-to-close baseline.