ResNet101 Chart Classification for A-Share Pullback Selection
Summary
The article describes a long-only Chinese equity strategy that uses a ResNet101 image classifier to identify strong stocks during pullbacks in an established uptrend. Its rationale combines buying a retracement in a strong trend, favoring stocks that outperform the market, and avoiding broad market timing. The model classifies chart images into two positive pattern types and a negative class; people then inspect candidate charts for a second feature. A 10% profit exit, other sell conditions, and a one-month maximum holding period guide trades.
The authors report tests across five market periods from 2014 to 2018, including bearish periods, and compare results before and after an additional three-part filter. They also describe optimizing capital allocation with a cash-flow planning model. These are historical results, not independent evidence of future performance. The article says the strategy could not be fully automated in a backtesting framework because candidate review remained manual. Its sample construction, manual decisions, execution assumptions, and limited test periods constrain reproducibility and confidence in the reported returns.
Key ideas
- The strategy seeks strong stocks that pull back during an established uptrend.
- A ResNet101 classifier reads chart images, while a human checks a second pattern feature.
- The reported exits include a profit target, other sell conditions, and a maximum holding period.
- A three-condition filter and a cash-allocation model are used to refine selection and capital use.
- Historical results are limited by manual screening and do not establish future performance.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.