Profit-sharing onboarding — selection criteria and Sharpe floor?
Could you clarify how the profit-sharing / performance-fee onboarding set is selected? A few specifics would really help me decide how to use my remaining submission slots:
- Roughly how many signals get onboarded — a fixed number, or all mutually-uncorrelated quality-set signals?
- Is there a Sharpe floor to qualify, or does any signal already in the quality set (significant positive Sharpe, <0.5 correlation) count?
- If a participant has multiple mutually-uncorrelated quality signals, are they all considered, or is it capped per participant?
- When is the onboarding selection finalized — before or after the Aug–Oct live simulation, and does live performance affect selection or only the fee share? Thanks for organizing this!
3 Replies
Nonius26d agoOnboarding & Selection Process Update
Thank you for the excellent questions regarding our onboarding framework. As you decide how to allocate your remaining submission slots, here is exactly how our selection process works:
#1. Selection Mechanism & Portfolio CapacityGreedy Selection: We use an algorithmic, greedy selection process.Incremental Integration: Signals are added one by one to our meta-strategy as long as they improve the aggregate portfolio.Performance Degradation: We stop onboarding once adding more signals causes the overall backtested performance to degrade from its peak.No Fixed Cap: There is no arbitrary limit; onboarding depends entirely on marginal portfolio value.
#2. Eligibility & Performance Floors
No Blanket Onboarding: Simply belonging to the quality set (significant Sharpe, low correlation) does not guarantee onboarding.
Marginal Value Rules: A signal must actively improve the net ensemble.
Turnover Penalty: We have observed multiple submissions with excessively high turnover. Under realistic execution assumptions, these degrade net performance due to transaction costs.
Upcoming Metric Update: Moving forward, we are implementing a strict transaction cost penalty directly into our Sharpe Ratio calculations to filter out high-turnover noise early.
#3. Multiple Submissions per Participant
All Submissions Evaluated: If you have multiple mutually-uncorrelated quality signals, all of them will be independently considered by our selection algorithm.
No Per-Participant Cap: We do not restrict how many models a single user can get deployed. If multiple signals from your portfolio add distinct value, we will onboard them all.
#4. Selection Timeline & Live Performance
No Signals Deployed Yet: To be transparent, we have not onboarded any signals to the live environment at this stage.
Finalization Timeline: Total Onboarding decisions will be finalized after the August–October live simulation phase. However , we may onboard a very select small set before the end of October .
Live Performance is Critical: Out-of-sample data is everything. We are closely monitoring recent months and the upcoming live simulation to see how signals hold up before making final deployment decisions. Live performance directly dictates if a signal gets selected, not just its fee share.Good luck with your final submissions! Let us know if you need any further clarifications.
About the transaction cost penalty mentioned in the profit sharing thread: will it apply to this competition's leaderboard and final scoring, or only to the onboarding selection, and from when?
Nonius21d agonext contest, not now. the signlas that have insanely large sharpes but are horrible under cost will not be integrated into our trading system. to be clear, some of the top signals currently are ok but still not hugely useful in the trading system; none are great on their own but if we blend a bunch of them it's decent. It's super hard trading on one signal, and mostly impossible most of the time. the value is in aggregating 100s of signals. the larger problem is mitigating cheaters in other dimensions but we have agents and the community flagging that.
Sign in to reply.