RollingInfoRatio

Description

Information ratio against a benchmark:

\[ \text{IR}[t] = \sqrt{\text{ppy}}\ \cdot\ \frac{\text{mean}(r - b)}{\text{std}(r - b)} \]

2-input, 1-output on (returns, benchmark). Effectively RollingSharpe applied to the active-return series r - b.

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 RollingInfoRatio

np.random.seed(0)
bench_ret = np.random.normal(0.0004, 0.012, size=300)
asset_ret = bench_ret + np.random.normal(0.0003, 0.006, size=300)
info = RollingInfoRatio(window_size=63)(asset_ret, bench_ret)   # Sharpe of active returns vs benchmark

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=asset_ret, mode="lines", name="asset returns"), row=1, col=1)
fig.add_trace(go.Scatter(y=bench_ret, mode="lines", name="benchmark returns"), row=1, col=1)
fig.add_trace(go.Scatter(y=info, mode="lines", name="info ratio",
                         line=dict(color="red")), row=2, col=1)
fig.update_layout(title="Rolling information ratio over 63 bars (RollingInfoRatio)",
                  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="info ratio", row=2, col=1)
fig.show()