returns_panel¶

Defined in fynance.research

returns_panel(experiments)[source]

Build a (T, n_configs) returns panel from research experiments.

One column per Experiment, taken from its stored series["returns"] (the per-period net returns run_experiment records). When an experiment carries only series["equity"] (e.g. a hand-built record), the column is derived as the equity curve’s percentage change, R_t = E_t / E_{t-1} - 1 for t = 1..T-1 — one observation shorter than the equity curve, since the first equity point has no preceding value to return from.

Parameters:

experimentslist of Experiment: The configurations to stack into a panel (>= 1).

Returns:

numpy.ndarray, shape (T, n_configs): Per-period returns, one config per column.

Raises:

ValueError: If experiments is empty, an experiment carries neither a "returns" nor an "equity" series (or an equity curve too short to derive a return from), or the resulting columns have mismatched lengths.

Examples

>>> from fynance.research import Experiment, returns_panel
>>> a = Experiment(name="a", series={"returns": [0.01, -0.02, 0.03]})
>>> b = Experiment(name="b", series={"equity": [100.0, 101.0, 99.0, 102.0]})
>>> returns_panel([a, b]).shape
(3, 2)