Hidden Markov Models for Chinese Stock Selection

Summary

This report adapts hidden Markov models (HMMs), a technique associated with speech recognition, to rank stocks by the likelihood of an upward move. It assumes that rising and falling stocks follow distinct patterns and models those patterns separately. Six price and volume indicators, including turnover and one-day returns, serve as observations; rising-stock examples train the model representing upward behavior.

For each stock, the strategy uses its observation probability under the upward-pattern model as a signal, sorts the universe into ten groups, and overweights the highest-ranked group. The report describes a backtest using CSI 500 constituents and says industry-neutral optimization improved the results. It reports annualized excess return, maximum drawdown, and information ratio for the stated test period. These are historical backtest findings, not evidence of future performance. The authors caution that market structure, trading behavior, or adoption by similar traders could weaken or invalidate the strategy.

Key ideas

• The strategy uses an HMM to represent patterns associated with rising stocks.
• Six price and volume indicators provide the model observations.
• Stocks are ranked by their likelihood under the upward-pattern model, and the top group is overweighted.
• The reported CSI 500 backtest improved after industry-neutral optimization.
• Changes in market behavior or wider adoption may cause the strategy to fail.

Tags

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