quantlab.model.predefined.smoothed#
Smooth a predictor’s predictions in time, causally.
A portfolio rule that trades every bar on the latest predictions trades
their fast part as well as their slow part. SmoothedPredictor wraps any
Predictor and replaces each bar’s prediction with an exponentially
weighted mean of the symbol’s own predictions up to and including that bar,
so a memoryless rule on it trades the slow part only. Bar t reads bars up to
t and nothing later.
Each bar is first standardised over its symbols (a cross-sectional z-score),
so a change of scale between bars, between the checkpoints of a walk-forward
run for example, does not weigh one stretch of bars more than another; the
smoothed panel is then on a standardised scale. standardize=False
smooths the raw values in their own units instead.
The mean at a window’s first bars needs the predictions of the bars before
it, the warm-up, which the wrapped predictor is asked for too. A model
predicts them with its loaded checkpoint; a RecordedPredictor with
history=True serves what the checkpoints live at those bars predicted
there (in a run_cv, the earlier folds), the history a live account
smoothing its own past predictions would hold.
Classes
A |
|
What a |