RollingSortino

Description

Annualised Sortino ratio:

\[ \text{Sortino}[t] = \sqrt{\text{ppy}}\ \cdot\ \frac{\text{mean}(r) - \text{target}}{\sqrt{\text{mean}(\min(r - \text{target},\ 0)^2)}} \]

Same as Sharpe but the denominator is the downside deviation -- only bars below target contribute, so upside variability is not penalised.

Implementation

O(window_size) per step. The downside-RMS denominator does not have a closed-form O(1) update.

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 RollingSortino

np.random.seed(0)
ret = np.random.normal(0.0005, 0.01, size=300)
sortino = RollingSortino(window_size=63, periods_per_year=252)(ret)   # Sharpe using only downside deviation

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=ret, mode="lines", name="returns"), row=1, col=1)
fig.add_trace(go.Scatter(y=sortino, mode="lines", name="Sortino",
                         line=dict(color="red")), row=2, col=1)
fig.update_layout(title="Annualised Sortino over 63 bars (RollingSortino)",
                  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="returns", row=1, col=1)
fig.update_yaxes(title_text="Sortino", row=2, col=1)
fig.show()