Virtual Regret Matching for Long–Short Index Timing

Summary

The article adapts virtual regret matching, a repeated-game method, to daily index timing. It first illustrates regret by comparing the payoff of an action taken with the payoff that would have resulted from alternatives. Cumulative regret for long and short actions is then used to update their execution probabilities. The proposed implementation tracks roughly one hundred days of CSI 300 closing prices, calculates daily returns and counterfactual action values, discounts accumulated regret with a factor of 0.9, and goes long when the long probability exceeds 0.55. Its initial long and short probabilities are 0.45 and 0.55.

The author reports that the backtest timing performance was not good and questions whether the regret calculations, including possible negative regret values and initialization, were interpreted correctly. The method is presented as an exploratory, locally adaptive alternative to conventional timing approaches, not as a validated strategy. Because it relies on a rolling history, it can lag market changes, and the article says implementation details need further research.

Key ideas

• Virtual regret measures the payoff forgone by not choosing an alternative action.
• Cumulative regret for long and short positions is used to update their action probabilities.
• The example applies the method to daily CSI 300 returns and discounts older regret values.
• The reported backtest performs poorly, and the author raises questions about implementation details.
• A rolling-history approach may respond slowly to changing market conditions.

Tags

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