RollingGarmanKlassVar

Description

The Garman-Klass (1980) range-based volatility estimator. Per-bar variance contribution:

\[ \sigma^2_\text{GK}[t] = \tfrac{1}{2}\big(\ln H/L\big)^2 - (2\ln 2 - 1)\big(\ln C/O\big)^2 \]

This expression is averaged with a rolling mean over window_size bars to form the estimator. The Vol variant returns sqrt(Var) (bit-exact via the same internal state).

4-input, 1-output on (open, high, low, close). ~7.4x more statistically efficient than close-to-close RollingStd under the model's assumptions (zero drift, no overnight gaps).

NaN handling

Policy: ignore. A NaN in any input at index t causes the function to skip that step: output at t is NaN and internal state is unchanged. Subsequent finite samples are processed as if step t had not occurred.

Examples

Usage example

import numpy as np
import plotly.graph_objects as go
from plotly.subplots import make_subplots
from screamer import RollingGarmanKlassVar

np.random.seed(0)
close = 100*np.exp(np.cumsum(np.random.normal(0, 0.01, size=300)))
open_ = np.concatenate([[close[0]], close[:-1]])
wick = np.abs(np.random.normal(0, 0.4, size=300))
high = np.maximum(open_, close) + wick
low  = np.minimum(open_, close) - wick
out = RollingGarmanKlassVar(window_size=20)(open_, high, low, close)

fig = make_subplots(rows=2, cols=1, shared_xaxes=True,
                    row_heights=[0.55, 0.45], vertical_spacing=0.08)
fig.add_trace(go.Scatter(y=close, name="close"), row=1, col=1)
fig.add_trace(go.Scatter(y=out, name="RollingGarmanKlassVar", line=dict(color="red")), row=2, col=1)
fig.update_layout(title="Rolling Garman-Klass variance (RollingGarmanKlassVar)",
                  margin=dict(l=20, r=20, t=60, b=20),
                  legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="right", x=1))
fig.update_yaxes(title_text="price", row=1, col=1)
fig.update_yaxes(title_text="variance", row=2, col=1)
fig.show()