Quantitative Investing in China: Reinforcement Learning, Explainability, and Competition
Summary
This overview of a Chinese securities firm’s research describes a shared AI platform for financial data, model development, GPU computing, backtesting, performance analysis, and deployment. It also discusses deep reinforcement learning for stock selection and trade execution. An execution strategy learned from A-share tick data to choose trade timing, size, and order type, with the reported live and historical tests comparing performance with market VWAP. The article does not provide enough methodological detail to independently assess those results.
The discussion emphasizes that model sophistication alone does not create an edge: data quality, prior knowledge, iteration, and strategy crowding matter. It presents explainability as a way to investigate overfitting, noisy or lucky factors, drawdowns, model boundaries, and changing market regimes. The final sections consider how quantitative strategies interact with other investors and how competition can erode excess returns. These claims are largely presented through interviews and selected examples; market structure and historical performance may differ across periods, and the article warns that models can lose effectiveness as conditions change.
Key ideas
- A shared platform can combine financial data, model research, computing, backtesting, and deployment resources.
- Deep reinforcement learning can support execution decisions about timing, quantity, and order type using tick data.
- An advanced algorithm cannot compensate for weak factors or poor data, and similar objectives can crowd strategies.
- Explainability can help investigate overfitting, drawdowns, noise, model limits, and shifts in market regimes.
- Competition among quantitative strategies can reduce excess returns and change the opportunities available to investors.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.