Rolling Windows and Feature Clipping in Quantitative Data

Summary

The document explains two settings in a rolling-data module used to prepare features for securities. The window size controls how many consecutive observations are grouped into each sample. Its example uses five sequential values and a window of three to illustrate overlapping windows, and it also describes packaging each security’s multi-feature history into a data segment for input and target pairs.

Feature clipping limits extreme factor values, typically after standardization, as a form of outlier handling. The explanation is informal and brief: it does not specify whether clipping applies symmetrically, how thresholds are chosen, or whether values are capped or removed. It provides no tests or evidence about the effect of either setting on model performance, so implementation details should be checked in the relevant module output.

Key ideas

• Window size determines the number of consecutive rows grouped into each sample.
• Rolling windows overlap as they advance through the observations.
• The module groups a security’s feature history into a segment for model inputs and targets.
• Feature clipping limits extreme factor values, often after standardization.
• The document does not specify clipping thresholds or exact handling of outliers.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.