Weekly Chinese Equity Timing with a Linear-Chain Conditional Random Field

Summary

The report describes a weekly market-timing model based on a linear-chain conditional random field (CRF). It frames the task as inferring market direction states from observed features, with CRFs able to represent dependencies between observations and successive states. The authors contrast this discriminative approach with hidden Markov models and emphasize choosing a compact model to limit parameter estimation problems and overfitting.

Feature development starts from returns and turnover, expands these into candidate variables, then removes redundant or ineffective inputs. The final set includes returns, return volatility, and sequences of consecutive rising or falling closes; the selected structure has three nodes. Reported tests cover five Chinese equity indices from late 2014 to early 2018, with the authors describing out-of-sample performance as generally good. For the CSI 300, annualized return is reported at about 14% before costs and 12% after a stated two-sided transaction cost. The summary provides limited methodological and evaluation detail, and the authors present the work as research rather than investment advice.

Key ideas

• The weekly timing method uses a linear-chain CRF to infer market states from observed indicators.
• Feature selection begins with returns and turnover, then retains returns, return volatility, and consecutive closing-price direction sequences.
• The report selects a three-node chain and cautions against excessive parameters because they can encourage overfitting.
• Tests span five Chinese equity indices from late 2014 to early 2018.
• The reported CSI 300 annualized return is about 14% before costs and 12% after the stated transaction cost.

Tags

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