How Survivorship Bias Inflated a China Small-Cap Backtest

Summary

The document examines a top-ranked strategy that selects the smallest stocks in the CSI 1000 and reports unusually strong historical performance. Its central lesson is that a backtest must use index constituents as they were known at each historical date. Using the index’s later constituent list in earlier periods introduces look-ahead and survivorship bias: stocks that grew into the index can appear in the historical portfolio before they qualified.

The author compares the strategy’s selections with results from other platforms and finds that the tested platform’s picks match the smallest stocks in a later CSI 1000 constituent list. A second platform’s version of the strategy loses money over the cited period, while the original reports a very large gain. The article attributes this gap to the constituent-data error, illustrating the issue with one stock that entered the index after its market value had grown. The evidence is a platform comparison and constituent check, not an independent full audit; the conclusion is specific to the examined strategy and data behavior.

Key ideas

• Historical index strategies need point-in-time constituent data to avoid look-ahead bias.
• Using current constituents in a historical backtest can favor companies that joined an index later.
• Different platforms can produce sharply different results when their historical data handling differs.
• The article tests its explanation by matching the backtest’s picks against a later constituent list.

Tags

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