Hour-Aware Adaptive Risk Management for Autonomous Memecoin Trading on Solana DEXs: Evidence, Theory, and Design Lessons from a 15-Day Deploymen

01 ABSTRACT

The study conducted 15 days of paper-traded autonomous memecoin trading on Solana DEXs, with 190 trades, a 40.5% win rate, mean per-trade return +0.62%, cumulative +117.7%, skewness -1.21, and excess kurtosis 6.61. Mann-Whitney U test between three exploratorily identified worst entry hours (mean -11.60%) and all others (mean +1.82%) yields p = 0.5634, directional but non-confirmatory. A counterfactual rejection tracker collected 4,874 forward-sample observations across 184 rejection events; of 48 events observed for at least six hours, 56.25% reached a 50% drawdown from reference, and removing the top three trades (1.6% of sample) flips cumulative return unprofitable. The author connects these findings to Kyle (1985) informed flow, Precup-Sutton-Singh (2000) off-policy evaluation, and Bailey-Lopez de Prado (2014) deflated Sharpe. The author's view is that the principal contribution is measurement infrastructure and three transferable design lessons; the factual part includes the above statistics and the public release of code and data.

02 KEY FINDINGS

1. 190 paper trades over 15 days yielded +117.7% cumulative return, but the return distribution is negatively skewed (-1.21) and heavy-tailed (excess kurtosis 6.61), turning unprofitable when the top three trades are removed.
2. The return difference between the three worst entry hours and other hours is not statistically significant (Mann-Whitney U p = 0.5634), marking an exploratory finding.
3. Counterfactual rejection tracking shows 56.25% of 48 events observed for at least six hours reached a 50% drawdown from reference, suggesting some filter decisions may have avoided downside.