Execution Assumptions Change Short-Horizon Sharpe Rankings: Evidence from a Synthetic Trading Benchmark
ORIGINAL / How Execution Assumptions Change Short-Horizon Sharpe Rankings: Evidence from a Synthetic Trading Benchmark
This study tests in a synthetic trading benchmark whether changing order-fill assumptions from idealized close prices to stressed settings with latency, spread, participation, and impact alters the short-horizon Sharpe rankings of LLM and classical baselines. It has direct methodological implications for avoiding strategy superiority claims based only on backtest rankings.
01 ABSTRACT
The author compares five prompted LLM signal policies and seven classical baselines on identical synthetic price paths under six execution settings, finding that in the high-volatility regime the near-ideal versus default-stress ranking agreement falls to Kendall τb=0.21, compared with 0.82 in the calm regime. Agreement rises as the asset board widens from two to ten, rank changes are related to turnover, and buy-and-hold comparisons depend on how the anchor is initialized. The author concludes that on short horizons an execution convention can change the ranking itself, so strategy comparisons should not report returns alone.
02 KEY FINDINGS
- Under high volatility, agreement between near-ideal fills and default-stress settings drops to τb=0.21 for short-horizon Sharpe rankings, versus 0.82 in calm conditions.
- Seed-bootstrap intervals are wide and overlapping, and the two-asset point estimate differs substantially from wider asset settings.
- Rank changes are related to turnover, and comparisons with buy-and-hold also depend on how that anchor is initialized.
AI GENERATED SUMMARY / DISCOVERED BY ARXIV CS.LG