PHASE 2 / EXECUTION ROBUSTNESS & PRE-LIVE
The strategy is research-frozen, not live-proven. For convertible bonds, execution quality is part of the evidence, not a detail after the evidence.
这条策略是研究封板，不是实盘已证明。对可转债来说，执行质量本身就是证据的一部分，不是证据之后的小细节。
可转债的吸引力来自“较小资金规模下仍可低频实施”，但这也要求研究不能停在漂亮回测上。候选标的是否具备可执行流动性，强赎数据是否及时，真实成本是否低于假设，调仓延迟是否会破坏路径，这些问题会直接决定策略是否仍然成立。
Convertible bonds are attractive partly because smaller capital can still implement the strategy at low frequency. That also means the research cannot stop at a nice backtest. Whether ranked candidates have executable liquidity, whether call-risk data is timely, whether realized cost stays below assumptions, and whether entry delay damages the path are all part of whether the strategy still works.
| Assumption | Why it is conservative | 中文说明 |
|---|---|---|
| Weekly rebalance | Keeps the process maintainable and avoids short-term noise chasing. | 周频调仓更适合个人维护，也避免追逐短期噪音。 |
| T+2 entry | Separates signal generation from review and trade execution. | 信号生成、人工核对和可实施执行之间留出延迟。 |
| 15 bps one-way cost | Forces the edge to survive non-trivial friction. | 用不算轻的成本口径提前压力测试。 |
| Strong-call PIT filter | Avoids using future call-risk knowledge. | 强赎剔除按时点可得处理，避免未来函数。 |
| Research freeze before pre-live | Stops parameter drift before paper trading. | 进入前向验证前先停止参数继续漂移。 |
这些假设不是为了显得保守，而是为了防止研究阶段把执行风险藏起来。低频策略如果只能在理想成交下成立，就不适合进入前向验证。
These assumptions are not cosmetic. They prevent execution risk from being hidden during research. If a low-frequency strategy works only under ideal fills, it is not ready for forward validation.
| Entry lag | top12_keep37 CAGR |
MDD | Calmar | Reading |
|---|---|---|---|---|
| T+1 | 20.34% | -22.39% | 0.908 | Stronger return, deeper drawdown. |
| T+2 | 18.60% | -18.76% | 0.992 | Main conservative convention. |
| T+3 | 17.29% | -24.97% | 0.692 | Still viable but visibly weaker. |
T+2 不是 bug，而是执行约定。T+1、T+2、T+3 都要看，是为了确认策略不是靠某一天理想成交撑起来。结果显示最终版在不同滞后下仍有韧性，但 T+3 明显变弱，所以执行纪律不能被忽略。
T+2 is not a bug; it is an execution convention. Testing T+1, T+2 and T+3 checks whether the strategy depends on one ideal fill day. The final candidate remains resilient under different lags, but T+3 weakens visibly, so execution discipline matters.