directional_accuracy¶

Defined in fynance.features.stats

directional_accuracy(y_true, y_pred, axis=0)[source]

Compute the directional accuracy of a prediction.

Fraction of periods where the predicted direction (sign) matches the true direction. A value of 1.0 means perfect directional alignment; 0.5 is random; 0.0 means systematically wrong direction.

Parameters:

y_truenp.ndarray[ndim=1 or 2, dtype]: Vector of true values (returns or price changes).
y_prednp.ndarray[ndim=1 or 2, dtype]: Vector of predicted values.
axis{0, 1}, optional: Axis along which the computation is done. Default is 0.

Returns:

float or np.ndarray[ndim=1, float]: Directional accuracy between 0 and 1.

See also

accuracy

Notes

\[directionalAccuracy = \frac{1}{T} \sum_{t=1}^{T} \mathbf{1}[\text{sign}(\hat{y}_t) = \text{sign}(y_t)]\]

Examples

>>> y_true = np.array([1., .5, -.5, .8, -.2])
>>> y_pred = np.array([.5, .2, -.5, .1, .0])
>>> directional_accuracy(y_true, y_pred)
0.8