RE:CZ

It is currently 4:00 AM on February 3, 2026.

I woke up extremely early today, not because I stayed up late. Yesterday, I used OpenCode + Opus 4.5 to help me complete the experimental work for the Capital Persistence War, and I had a sudden sense of clarity. The efficiency improvement brought by AI is truly a qualitative leap. I can now direct my AI assistants to complete various experimental tasks like a lab supervisor, and then I analyze the results, propose new hypotheses, and design new experiments.

Through yesterday's experiments, I gained a deeper understanding of the experimental design for the Capital Persistence War. The redesign of the experiment allows us to more systematically evaluate the performance of different signal strategies and betting strategies across various market environments. In the future, I will develop this into a quantitative strategy incubator, capable of rapidly testing and iterating various strategy combinations, ultimately deploying them into real trading environments.

Market sequences exist objectively, signal strategies are reactions to the market, and betting strategies reflect the risk preferences and capital management methods of the investment entity. By combining these three, we can more comprehensively understand the effectiveness of an investment strategy. In the experiments, overall, the Signal-Market Fit (SMF)¹ between the signal strategy and the market sequence determines the performance of the benchmark account, while the betting strategy determines the efficiency of capital management and the effectiveness of risk control, ultimately determining the actual utility for the investment entity.

How Should We View Human Subjective Trading?

This leads me to a new line of thought: How should we view human subjective trading?

As mentioned in the Cathay Haitong 2025 Annual Report, the organic combination of subjective and quantitative trading is a future trend.

My view is: Subjective trading should only act on signal strategies and must never interfere with betting strategies. Human subjective trading ability is more reflected in understanding market environments and designing signal strategies, not in capital management and risk control. The greatest weakness of humans lies in irrational decision-making in capital management. Faced with losses, they tend to become fearful; faced with profits, they tend to become greedy. If we can completely delegate betting strategies to programs for execution, we can avoid the emotional issues humans have in capital management. Humans must be unable to intervene or even remain completely unaware of the betting strategy.

Of course, it's impossible to completely prevent humans from peeking at the performance of the betting account. If they do, it will affect the mindset of the subjective trader, thereby influencing the design of the signal strategy. The key flaw here is that humans can discover the strong causal relationship between their own signal strategy and the betting account. To avoid this problem, we must find a way to sever the path through which humans discover this strong causal link.

Community-Based Trading

Have multiple subjective traders act as different signal strategies. Subjective traders can only see the performance of their own signal strategy and the performance of the consolidated betting account.

This way, the performance of the consolidated betting account no longer has a strong causal relationship with the performance of any single signal strategy, thereby avoiding the possibility of subjective traders adjusting their signal strategies based on the betting account's performance. This approach creates an effect similar to collective wisdom, better capturing market diversity. Moreover, due to the capital management characteristics of the anti-Martingale strategy, extreme losses from a single signal strategy will not significantly impact the overall betting account. Risk is also controlled, as it can tolerate the failure of individual signal strategies while actively responding to successful ones.

Community participants can only invest funds into the consolidated betting account fund and cannot directly intervene in the capital management of any single signal strategy. The profits of the consolidated betting fund are distributed 80% to capital providers and 20% to the designers of the signal strategies (subjective traders). For performance attribution, rewards can be allocated post-hoc based on each signal strategy's contribution to the consolidated betting account. Here, a trade-off between fairness and efficiency can be made, ultimately aiming to encourage consistency and effectiveness in signal strategies. In simple terms, it encourages subjective traders to design better signal strategies and execute them with undivided attention.

Efficiency means that profits should be allocated entirely according to the contribution of the signal strategy—those who contribute more receive more. Fairness means that even if a signal strategy's contribution appears small in hindsight, as long as it meets basic quality standards, it should receive a certain reward to encourage more participants to join the community.

Pursuing absolute efficiency could lead to systemic arbitrage, such as guiding a trader to hedge with multiple signals and then obtaining huge profits through profitable signals. This doesn't actually generate any new signal strategy value; it merely exploits loopholes in the rules.

Pursuing absolute fairness could lead to a decline in overall returns because rewards are distributed too evenly, failing to incentivize traders to design better signal strategies. Excellent signal strategy designers might lose motivation if their profits are diluted.