Rolling Train-Test Evaluation with Nonoverlapping Time Periods

Summary

The discussion describes a proposed rolling evaluation in which training and testing periods are kept separate. Its example uses successive annual training windows and test windows offset into later years, with each test period’s results concatenated to form an overall record. The approach is intended to assess performance across multiple time shifts while avoiding overlap between the training data and the corresponding test data.

The post is a request for implementation guidance rather than a worked solution. It suggests that a custom execution module might support the setup, but offers no code, platform-specific procedure, or validation of that suggestion. It also does not explain choices such as window length, model refitting, or safeguards against look-ahead bias, so those details would need to be specified before treating the evaluation as a reliable backtest.

Key ideas

• The proposed evaluation keeps each training period separate from its corresponding test period.
• Training and test windows advance through time in annual steps.
• The author proposes joining the successive test results into one record.
• The post asks how to implement the scheme and does not provide a solution.

Tags

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