A critical engineering post from a Chinese quant developer exposes a subtle but devastating data leakage problem in A-share backtesting. The issue stems from using forward-adjusted prices (前复权), which incorporate future corporate actions like stock splits and dividends into historical price data. This creates a Point-in-Time bias: the backtest sees information that was not available at the time of trading, artificially inflating performance metrics like annualized returns. The author systematically compares unadjusted, forward-adjusted, and backward-adjusted price methods, deriving linear relationships and providing Python code for validation and mapping between backtest and live trading prices. For quantitative developers and financial engineers, this is a foundational lesson in data integrity. The post is not a tutorial but a real-world engineering case study, making it highly valuable for building robust backtesting frameworks. It underscores the need for point-in-time data management, a practice often overlooked in retail and even some professional quant setups. The commercial value is high: avoiding this bias can save teams from deploying flawed strategies based on backtest results that cannot be replicated live.
A deep dive into how forward-adjusted prices cause Point-in-Time bias in A-share backtesting, leading to data leakage and inflated returns.