ML sizing boosted returns claim
A recent social post claims ML‑driven conviction sizing lifted a baseline strategy’s returns by 83%, achieved a ~70% win rate, and cut drawdown ~46% — a high‑engagement example of conviction‑based sizing buzz on X. The post is anecdotal but underscores the appetite for position‑sizing experiments that pair simple alphas with adaptive risk managers. (x.com)
The X post making the conviction‑sizing claim did not publish backtest code, a parameter list, or an explicit in‑sample/out‑of‑sample split, information that prevents independent validation of the claim. (x.com) (x.com) Academic and industry backtesting studies show that strategies with spectacular in‑sample performance often collapse out of sample unless they report walk‑forward validation, purged k‑fold cross‑validation, or an explicit out‑of‑sample holdout; one PM Research review and Quant trading guides document large disparities between in‑sample and true out‑of‑sample performance. (pm-research.com) (pm-research.com; paperswithbacktest.com) Win‑rate statistics are sensitive to sample size and streak variability: probability tables used by trading educators demonstrate that even genuine edges can produce long losing streaks and that high win rates can arise from small or selected samples. (newtraderu.com) (newtraderu.com; quantifiedstrategies.com) “Conviction” sizing schemes often amount to adaptive Kelly or anti‑Martingale scaling; Kelly‑type sizing can mathematically maximize growth under fixed win/loss distributions but requires accurate edge and payoff estimates that many anecdotal posts omit. (tradesearcher.ai) (tradesearcher.ai; hyper-quant.tech) Transaction costs, slippage and market impact routinely turn modelled excess returns into losses unless explicitly modelled: institutional guides recommend per‑market cost assumptions and show round‑trip frictions of 0.25–0.75% or higher can eliminate simulated edges. (backtestme.com) (backtestme.com; bsic.it) Robust validation steps for a conviction‑sizing claim are concrete and testable: report the rolling walk‑forward Sharpe, turnover, max drawdown, a capacity estimate, and apply purged/Combinatorial Purged Cross‑Validation before any live allocation. (quantconnect.com) (quantconnect.com; paperswithbacktest.com) A practical final check used in quant shops is paper trading with live slippage and fills for at least one full market regime (e.g., three to twelve months depending on strategy turnover) before risking capital, a recommendation echoed in quantified backtesting guides. (quantifiedstrategies.com) (quantifiedstrategies.com; quantconnect.com)
