CNN Stock Selection with Momentum, Money Flow, and Trend Filters

Summary

The strategy turns conventional stock-selection rules into Boolean features for a CNN classifier. Its inputs include returns and return ranks, turnover, valuation and profitability measures, money-flow signals, volatility, technical indicators, and recent price-limit events. Selection filters favor strong short-term trends and positive main-fund flows, require bullish moving-average and MACD conditions, exclude special-treatment shares, and keep the session low above the 20-day average. The stated intent is to identify stocks with near-term strength and potential for rapid gains.

For model labels, the article proposes comparing the highest price over the next three days with the next day's opening price, arguing that this captures intraday upside missed by closing-price labels. It also describes filtering for large-order inflows and suggests tuning the rolling training window, replacing stale or crowded factors, and refreshing the training period. The article reports that some money-flow factor families appeared useful in single-factor tests, but supplies no detailed CNN validation or performance statistics. The forward-looking label also requires careful handling to avoid information leakage in training and evaluation.

Key ideas

• The strategy encodes hand-built stock-selection rules as Boolean features for a CNN classifier.
• Features span momentum, valuation, profitability, money flow, volatility, technical indicators, and price-limit history.
• Entry filters favor bullish trend conditions, positive large-order flow, and prices holding above the 20-day average.
• The proposed label uses the next three days' high relative to the next day's open to represent upside potential.
• The article gives no detailed model results and does not explain how it prevents future data from leaking into evaluation.

Tags

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