Numerai Model Analytics

Classic model participation, concentration, explicit scores and unannualized returns on settled positions. Every positive fractional stake counts; benchmarks are excluded from participant cohorts.

Model staking participation

Weekly counts of entrants, conversions, unstakers and prediction-only models. These are model-state transitions, not deposits or account exits.

Stake shares and concentration

History for today’s top seven model cohort, top-N shares, and 1/HHI concentration equivalents. Retrospective selection is disclosed; effective N does not measure independent strategies.

Stake and MMC60

Current round principal versus trailing final MMC60; matched-population equal and stake weighted means; final benchmark_blender comparisons; and within-regime quintile persistence with missing models shown separately. Association is not proof of skill or causal selection.

Settled model outcomes

Share of eligible models with positive net NMR and distribution of mean position returns over the trailing 52 eligible weekly rounds, with at least 13 settled observations. Windows restart at a regime change. Mean return divides net NMR by summed position principal, without multiplying by observation count.

Return per settled round

Net NMR divided by the principal of those same settled model positions. No annualization or axis clipping. Atomic and legacy position denominators differ.

Are strong contributors staying?

A fixed cohort with twelve months of follow-up compares retention of top-quarter prior MMC60 models with other scored models. Groups use thirteen weekly score rounds opened at least one hundred days before the cohort. Monthly checkpoints count positive stake, including returns after gaps. Historical score revisions are not fully archived; model slots can change strategy.

Model survival and age

Positive-stake participation by first observed year, with recent activity censored, and stake-weighted model-slot age. A model can change strategy or replace another model from the same account.

Score-sign distributions

Positive versus non-positive final MMC60 and CORR60 among selected top staked models. Score signs do not determine combined payout signs.

Estimated Classic stake allocation

Weekly peak of legacy stake ×1 and atomic round principal ×64 from round 1343. The atomic estimate assumes 64 overlapping rounds; it is not measured locked capital. Latest week provisional.