小时感知的自适应风险管理用于自主Memecoin交易：一个多层智能框架
Hour-Aware Adaptive Risk Management for Autonomous Memecoin Trading on Solana DEXs: Evidence, Theory, and Design Lessons from a 15-Day Deployment
AI总结：
本文在Solana去中心化交易所上部署自主memecoin交易系统，通过190笔交易样本（40.5%胜率，累计+117.7%）评估小时效应、过滤器精度、脆弱性和已实现收益，发现盈利能力结构脆弱。
摘要：
本文在Solana去中心化交易所上对一个自主memecoin交易系统进行了为期15天的纸交易部署，测量了小时效应、过滤器精度、脆弱性和已实现收益。190笔交易样本（2026年3月29日至4月12日）显示胜率40.5%，每笔交易平均回报+0.62%，累计+117.7%（净SOL +0.039），偏度-1.21，超额峰度6.61。对三个表现最差的UTC小时（2、13、23）与其他小时进行Mann-Whitney U检验，得到U = 1,274，p = 0.22；在n=190时具有方向性但不显著。这三个小时是在样本内选择的，因此比较是探索性的，而非验证性的。一个并行的反事实拒绝追踪系统收集了184个不同拒绝事件中的4,874个前向样本观测值。在这些事件中，17.9%在24小时内从参考点回撤达到50%；26.0%的前向样本记录被拒绝的代币低于半参考点。过滤器堆栈避免了这些已实现回撤：证据表明拒绝标准相对于前向市场结果是净正向的。脆弱性是主要的警告。移除前三笔交易（样本的1.6%）会使累计回报变为亏损。盈利能力依赖于少数大赢家，并且结构上脆弱。数据集和审计脚本已根据CC-BY-4.0存放（Zenodo DOI 10.5281/zenodo.20043302）。
英文摘要：
We report a 15-day paper-traded autonomous memecoin trading deployment on Solana decentralised exchanges (DEXs), designed as a controlled measurement of three microstructure questions on which classical equity theory offers well-defined predictions but on which the AMM Solana venue lacks published measurement: (i) time-of-day return patterns on a 24/7 permissionless venue; (ii) whether decision-time filter stacks are net-positive against counterfactual returns of rejected tokens; (iii) whether small-sample cumulative-return statistics on a heavy-tailed venue are structurally robust or fragile. The 190-trade sample (March 29 to April 12, 2026) shows a 40.5 percent win rate, mean per-trade return +0.62 percent, cumulative +117.7 percent, skewness -1.21, excess kurtosis 6.61. Mann-Whitney U on three exploratorily identified worst entry hours (n=17, mean -11.60 percent) versus all others (n=173, mean +1.82 percent) yields p = 0.5634; directional and non-confirmatory. A parallel counterfactual rejection-tracker collected 4,874 forward-sample observations across 184 rejection events; of 48 events observed for at least six hours, 27 (56.25 percent) reached a 50 percent drawdown from reference (the full-cohort 17.9 percent is a censored lower bound). Removing the top three trades (1.6 percent of sample) flips cumulative return unprofitable. The three findings connect to Kyle (1985) informed-flow, Precup-Sutton-Singh (2000) off-policy evaluation, and Bailey-Lopez de Prado (2014) deflated-Sharpe predictions. Alongside the trade log and rejection-sample corpus (CC-BY-4.0), we deposit audit.py (MIT), an assertion-based reproduction script that exits zero iff every headline number reproduces from the deposited CSVs. Companion dataset: Zenodo concept DOI 10.5281/zenodo.20043301. The paper's principal contribution is measurement infrastructure and three transferable design lessons.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
