RollingMaxDrawdown

Description

The worst peak-to-trough loss observed inside the last window_size bars. Different from MaxDrawdown (which is the worst loss EVER since reset).

Implementation

Maintains a circular buffer of the last w prices and, on each step, sweeps the buffer tracking a within-window running peak and the worst drawdown from that peak. O(window_size) per step -- there is no cheap amortised algorithm for the standard definition because the in-window peak can sit anywhere in the window.

If you want the cheaper "current drawdown vs. rolling-window peak" approximation, compose it directly:

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 RollingMaxDrawdown

np.random.seed(0)
price = 100 * np.exp(np.cumsum(np.random.normal(0.0005, 0.02, size=300)))
rmdd = RollingMaxDrawdown(window_size=63)(price)   # worst drawdown in the last 63 bars

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=price, mode="lines", name="price"), row=1, col=1)
fig.add_trace(go.Scatter(y=rmdd, mode="lines", name="rolling max drawdown",
                         line=dict(color="red"), fill="tozeroy"), row=2, col=1)
fig.update_layout(title="Worst drawdown in a trailing window (RollingMaxDrawdown)",
                  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="rolling max drawdown", row=2, col=1)
fig.show()