Hybrid Kelly Sizing with Prop Firm Risk Limits and CPCV Validation
Summary
The article compares Kelly sizing with a confidence based signal method, then proposes combining them in two stages: the signal method accounts for estimated confidence and overlapping positions, while a Kelly multiplier adjusts for asymmetric win and loss sizes. It explains that Kelly maximizes long run log growth under idealized repeated bets with known probabilities and payoffs, and gives a symmetric payoff comparison in which Kelly is more aggressive below a crossover near 0.88. Its main practical case for Kelly is incorporating a characterized win/loss ratio.
The proposed risk layer adjusts sizing to a prop firm account’s remaining daily and overall drawdown capacity. For evaluation, the article recommends simulating a fresh account state bar by bar over combinatorial cross validation paths, summarizing the range of equity outcomes and auditing probability of backtest overfitting. It emphasizes important limits: financial bets are dependent, probabilities and payoff ratios are estimated and can drift, and full Kelly can over-size under estimation error. The methods are presented as a framework, with the account rules and historical estimates requiring careful calibration and validation.
Key ideas
- Kelly sizing incorporates both estimated win probability and the payoff ratio.
- Kelly can be more aggressive than confidence based sizing in the probability range common to financial classifiers.
- A hybrid can preserve concurrency corrections while adjusting exposure for payoff asymmetry.
- Sizing can be scaled dynamically according to remaining account drawdown capacity.
- Combinatorial paths can expose how path-dependent sizing changes backtest outcomes.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.