CNN Stock Selection with Momentum, Technical, and Capital-Flow Features

Summary

This Chinese strategy note describes turning discretionary stock-screening rules into Boolean model features for a CNN classifier. The proposed universe favors companies with positive profitability and valuation measures, short-term strength, positive main-fund flows, and recent limit-up activity. Candidate features include returns, turnover, market capitalization, valuation, capital-flow measures, moving averages, volatility, momentum, and breakout conditions. The note reports that single-factor tests found some usefulness in four categories of capital-flow factors, but gives no detailed test results or performance statistics.

Its filters favor bullish DMA, MACD, and five-day moving-average states, require the low to remain above the 20-day average, and exclude special-treatment stocks. The training label uses the highest price over the next three days relative to the following day’s open, intended to capture upside missed by closing-price labels. A large-order inflow threshold is also suggested. The author recommends retuning rolling-window length, features, labels, screening rules, and training data. The note provides a strategy outline rather than full implementation or validated live-performance evidence.

Key ideas

• The strategy converts conventional screening rules into Boolean features for CNN classification.
• It combines financial, trend, breakout, turnover, and capital-flow signals to identify strong stocks.
• The proposed price label uses a future high relative to a future open to represent potential upside.
• The note cites single-factor findings for flow-related features but supplies no detailed results or robust performance evidence.
• Feature selection, labeling, filters, and training windows are presented as areas for further optimization.

Tags

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