MMC vs CORJ60: What Numerai Actually Pays For

Crossing from the bottom MMC decile into the middle of the field flips a model's average per-round payout from roughly -0.3 NMR to roughly +0.2 NMR, bigger than any swing CORJ60 produces on its own. That sign-flip is the most important fact about Numerai's payout formula: 0.5 × CORJ60 + 2.0 × MMC is not a tweak, it is a different game from raw correlation. Optimize for only one half and you may leave NMR on the table, or hand a 25% burn to the protocol.

Which metric actually pays better? The answer depends on whether you care about expected payout sign or tail outcomes.

What the Metrics Measure

CORJ60 is Numerai's 60-day correlation score, measuring how well your predictions align with actual market outcomes over that window. High CORJ60 means your model predicts returns well. The distribution across staked models clusters near zero with a slight positive skew.

MMC (Meta-Model Contribution) is subtler. It asks whether your model still adds value after subtracting the crowd's combined prediction (the meta-model). A high-MMC model knows something the rest of the tournament does not. You can post strong correlation and near-zero MMC if your predictions are redundant with everyone else's.

CORJ60 rewards being right. MMC rewards being right in a way nobody else is. Both metrics are charted over time on the trends page, with per-round distributions on any round detail view.

The Payout Formula

Numerai's payout formula makes its priorities explicit:

The MMC coefficient is four times the CORJ60 coefficient, though realized impact also depends on each metric's distribution. Numerai is a hedge fund, and redundant predictions do not help it trade. The meta-model needs diverse, uncorrelated signals, so the payout formula pays a steep premium for originality.

A model with mediocre correlation but strong MMC can outperform a high-correlation model that tracks the meta-model closely. Does that play out in the data?

MMC vs CORJ60: The Scatter View

The scatter plots each model-round observation with CORJ60 on the x-axis (about -0.08 to 0.13), MMC on the y-axis (about -0.06 to 0.07), and color encoding NMR payout. CORJ60 has a wider spread than MMC: roughly 0.2 across versus 0.13. Raw correlation varies more model-to-model than meta-model contribution does.

The two metrics are only weakly correlated. Plenty of models sit at high CORJ60 with near-zero MMC (accurate but unoriginal), and a smaller group lives at modest CORJ60 with positive MMC (contributing something new). The highest-payout observations (greenest points) cluster in the upper right, where both MMC and CORJ60 are positive. Pure contrarianism in the upper-left (high MMC, negative CORJ60) is thinly populated. Models that contribute uniquely to the meta-model tend to also clear a baseline level of accuracy.

Do Big Stakers Optimize for MMC?

This chart compares stake-weighted averages against simple medians for both MMC and CORJ60 across rounds 200–1200. The same comparison is available interactively via metric distribution on the models page. The gap between the two lines shows how capital is allocated relative to skill.

The two metrics behave very differently here. Stake-weighted CORJ60 tracks the median CORJ60 closely (both swing between roughly -0.02 and +0.06 with market regime), with the stake-weighted line typically sitting a touch above. For MMC the picture inverts: the median MMC oscillates in a narrow band around zero, while stake-weighted MMC is nearly flat at zero the entire time. Capital is slightly more accurate than the typical submission on CORJ60, and much steadier, not visibly higher, on MMC: the big stakes sit in models whose meta-model contribution barely wobbles round to round.

As stake flows toward high-MMC models, the meta-model absorbs more of their signal, raising the originality bar over time.

Payout Distribution: Top-10% MMC vs Top-10% CORJ60

To make the comparison concrete, we isolated the top 10% of models by MMC and the top 10% by CORJ60 in each round, then plotted their NMR payout distributions. The chart clips outliers beyond 10.3 NMR to keep the bulk readable.

The two violins look more alike than a payout-formula reading would predict. Both cohorts pile up within a fraction of an NMR of break-even, each with a thin spike of extreme rounds running past the clip. The visible difference is in the body: the top-CORJ60 cohort is a little wider at zero, while the top-MMC cohort carries slightly more mass in the 0.2–0.5 NMR band.

Elite metric scores, on either axis, mostly do not translate into large per-round paydays: payout magnitude depends on stake size as much as score, and most top-decile observations belong to small stakes.

Neither metric alone buys the right tail. Blending both (a model that clears the MMC bar while keeping real CORJ60) is how you capture the 2x MMC multiplier without giving up correlation's ceiling.

Is the MMC-Payout Relationship Linear?

Bucketing models into MMC deciles and charting average payout per decile makes the MMC-to-pay relationship direct. The bottom three deciles (0–30%) average negative payouts, bottoming out near -0.3 NMR. From the 30–40% decile onward the climb is steady, reaching about 0.6 NMR in the top decile.

Crossing from a bottom-quintile to a mid-range MMC flips the sign of your expected payout, the single biggest jump in the chart. After that, the climb is lumpy but averages roughly 0.1 NMR per decile. The largest payout improvement is moving out of the bottom MMC deciles; gains after that are more incremental.

Strategy Implications

A few takeaways from the data:

• MMC flips the sign of your expected payout. The bottom three MMC deciles lose money on average; everything above that earns. Clearing the bottom 30% is the single most important line to cross.
• Elite scores alone rarely mean big paydays. Both the top-MMC and top-CORJ60 cohorts cluster within a fraction of an NMR of break-even per round. Payout magnitude comes from score and stake size together.
• Stake is concentrated in steady, accurate models. Stake-weighted CORJ60 runs slightly above the field median, and stake-weighted MMC is far smoother than the median. You are not competing against the median submission, you are competing against the capital-weighted one.
• Originality compounds. As stake flows toward high-MMC models, the meta-model absorbs their signal, raising the bar for newcomers. Early movers on novel features and architectures have a structural advantage, and the benchmark models show how far a well-tuned public baseline can go.

The tournament's incentive structure is unusually transparent: the formula is public, the data is available on every round page, and the outcomes are observable. Models with weak MMC tend to transfer payout opportunity to models with differentiated positive signal.