Forecast-to-Fill: Benchmark-Neutral Alpha and Billion-Dollar Capacity in Gold Futures (2015–2025)

Battle Count: 70

This paper is directly and highly relevant to quantitative trading. It presents a complete, implementable trading system from signal generation to execution with explicit cost modeling, capacity analysis, and rigorous out-of-sample validation. The walk-forward methodology, friction-adjusted Kelly sizing, volatility targeting, and ATR-based exits are all core quantitative trading engineering components. The capacity frontier analysis (up to ~$1B AUM) and benchmark-neutral alpha (beta=0.03) are directly relevant to institutional allocator decisions. The paper addresses the critical gap between backtest performance and live tradability.

The strategy is intentionally designed to be simple and interpretable. Core components include: EMA smoothing (single parameter lambda), z-score standardization, linear probability mapping, 50-day momentum check, EWMA volatility forecast, quadratic Kelly optimization (closed-form solution), ATR calculation, and simple threshold-based entry/exit rules. No machine learning, no complex optimization solvers, no high-frequency infrastructure required. The main complexity lies in the walk-forward framework management, parameter freezing discipline, and capacity curve computation. All formulas are closed-form or recursive. A competent quantitative developer could implement this in 1-2 weeks.

The paper provides a detailed reproducibility checklist: (i) freeze all train-time parameters before each OOS slice, (ii) compute recursive filters forward-only, (iii) use T+1 fills in baseline, (iv) apply costs to realized |delta_w_t|, (v) log full decisions, fills, and costs. All parameters are explicitly defined with formulas. The paper mentions code/data release for participation distribution. However, no GitHub repository URL is provided, and the exact LBMA/COMEX data access details are not fully specified. The walk-forward methodology is fully described with all hyperparameters (lambda, omega=0.6, K=50, ATR_14, thresholds, lambda_Kelly=0.40, W_max=2.0, sigma_target=15%). SPA test parameters (64 configs, B=800, block=20) are disclosed.