RollingAlpha

Description

Companion to RollingBeta. The regression intercept of y on x:

\[ \alpha[t] = \overline{y}_w - \beta[t]\ \cdot\ \overline{x}_w \]

2-input, 1-output on (target, regressor) -- same convention as RollingBeta. Composes RollingBeta + two RollingMean instances. O(1) per step.

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 RollingAlpha

np.random.seed(0)
market_ret = np.random.normal(0.0004, 0.012, size=300)
asset_ret = 0.6 * market_ret + np.random.normal(0.0003, 0.008, size=300)
alpha = RollingAlpha(window_size=63)(asset_ret, market_ret)   # regression intercept of asset on market

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=market_ret, mode="lines", name="market returns"), row=1, col=1)
fig.add_trace(go.Scatter(y=alpha, mode="lines", name="alpha",
                         line=dict(color="red")), row=2, col=1)
fig.update_layout(title="Rolling alpha of asset on market over 63 bars (RollingAlpha)",
                  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="alpha", row=2, col=1)
fig.show()