Building a Chinese Stock Ranking Model and Equal-Weight Portfolio

Summary

This BigQuant workflow illustrates a daily Chinese equity ranking process. It prepares stock features from price and prefactor data, including lagged returns and volume ranking, then creates a supervised label from a clipped forward return. Training data is extracted for 2021–2023, with a separate prediction period from 2024 into 2025. A stock-ranking model produces scores, and the portfolio module selects ten names and assigns equal weights.

The trading engine checks for scheduled signal dates, exits holdings outside the target list, and sets target weights for selected stocks. The configuration specifies trading every five sessions and uses opening prices for orders. These are implementation settings rather than evidence of performance: the document provides no reported returns, benchmark comparison, or risk statistics. The feature and label descriptions also do not establish that the workflow avoids look-ahead bias or reflects realistic trading costs and constraints, so the example needs validation before drawing conclusions about its merits.

Key ideas

• The workflow derives stock features from historical price and prefactor data.
• It labels training examples using clipped forward returns and trains a ranking model.
• The model selects ten stocks for an equal-weight portfolio.
• The trading engine rebalances on a five-session schedule and targets opening-price orders.
• No performance results or risk analysis are provided, so the workflow's effectiveness is not established.

Tags

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