`ROCR`#

Description#

ROCR(k) is the rate of change over k steps, expressed as a ratio:

\[ \text{ROCR}[t] = \frac{x[t]}{x[t-k]} \]

The ratio form is convenient when you want to chain returns multiplicatively (e.g. cumulative-product wealth paths) without subtracting 1 each step.

Parameters:

window_size (int, positive): the lookback k.

NaN handling: NaN for the first k samples; NaN when x[t-k] == 0.

Identity to ROCP#

ROCR(k) - 1 == ROCP(k) exactly. Pick whichever form keeps the calling code cleaner.

NaN handling#

Policy: propagate. Input NaN values are stored in the lookback. Output is NaN at any index where the function's positional formula references a NaN input; recovery happens once the NaN slides out of the lookback.

Examples#

Usage example#

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

np.random.seed(0)
price = 100 * np.exp(np.cumsum(np.random.normal(0.0005, 0.02, size=300)))
rocr = ROCR(window_size=20)(price)      # price ratio versus 20 bars ago, 1.0 = unchanged

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=rocr, mode="lines", name="ROCR",
                         line=dict(color="red")), row=2, col=1)
fig.update_layout(title="Price ratio over 20 bars (ROCR)",
                  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="ratio", row=2, col=1)
fig.show()

Reference#

Equivalent to talib.ROCR(x, timeperiod=k). Bit-exact match (cross-validated in tests/test_third_party_alignment.py).

ROCR#