Stock Scorer Shows Negative Predictive Power in A-Share Backtest

A developer built a three-pillar scoring model for Chinese A-shares, combining technical, volume/price, and valuation signals to rank stocks and select the top 50. When audited against 70 rebalance dates and 587 stocks, the composite score showed a significantly negative rank information coefficient, meaning higher scores tended to precede lower returns. Two of the three pillars had negative IC values, and the only positive one came from the component labeled fundamental, which the author suspects is mislabeled. Decile analysis reinforced the finding, as the highest-scoring stocks underperformed the lowest-scoring ones over the five-day horizon. The author used the same shipped code in the audit rather than a rewritten version, and noted that the results are not perfectly monotonic and occurred during a falling market window.