Is Staking on Numerai Profitable? A Data Look

Numerai lets anyone stake NMR on their model's predictions. Stake well and you earn more NMR. Stake poorly and your NMR is burned — permanently destroyed. How does that play out across roughly 1,200 rounds of real data?

Short answer: across all participants, earned NMR outpaces burned by about 1.4 million, so the system is net positive. But the median per-round return is only 0.05% of stake. Staking is profitable in aggregate and marginal in the middle. For a primer on the tournament itself, see How Numerai Works.

How Staking Works (Quick Refresher)

You can stake as little as 0.01 NMR on any model you control. Each round, your payout or burn is determined by your model's performance on two key metrics:

Payout = stake x (0.5 x CORR + 2.0 x MMC) x payout_factor

The maximum payout or burn per round is capped at 25% of your stake. If you stake 100 NMR and your model scores well, you might earn up to 25 NMR. If it scores poorly, you could lose up to 25 NMR — and that NMR is gone forever. It is not redistributed to other participants. It is burned on-chain.

The payout factor is a tournament-wide multiplier that Numerai adjusts over time. It modulates how much everyone earns or loses in a given round, and it has a significant impact on long-term profitability. More on that below, and see Round Economics for how it interacts with stake levels.

Cumulative Earned vs Burned

The most direct way to gauge tournament economics is to look at the aggregate NMR earned and burned across all participants over time.

By the latest round, cumulative earned sits near 2.5M NMR while cumulative burned is around 1.1M NMR, leaving a net of roughly 1.4M NMR paid out to stakers. In aggregate the tournament pays out more than it destroys. The same series is tracked live on the dashboard home and broken down round-by-round on the trends page.

The gap is not smooth. During rounds where model consensus breaks down, burn rates spike and the net curve flattens. Aggregate numbers also hide the individual distribution — a net-positive pool does not guarantee a net-positive staker.

Payout by Stake Tier

Do bigger stakers earn more per round, and is it just because they have more at risk?

Median per-round payout scales almost linearly with stake size. The 1K+ tier earns a median of ~1.3 NMR per round (n=112,227 round-model observations), the 100–1K tier earns ~0.1 NMR, and the three smaller tiers (10–100, 1–10, <1 NMR) sit essentially at zero. That is what you would expect if percentage returns are roughly similar across tiers: multiply a small percentage by a bigger base and the absolute number grows proportionally.

The sample-size labels also matter. Sub-1 NMR stakes account for 2.24M observations but almost no aggregate payout, while the 1K+ tier produces most of the visible NMR flows with far fewer observations. Big stakers move most of the tournament's NMR, but per unit staked they do not earn a dramatic premium. For a live breakdown see payout ROI by stake tier.

For new participants, that is encouraging: starting small does not bury you in a structurally worse payout regime.

Payout Factor Trends

The payout factor is arguably the single most important variable for staking profitability that you cannot control. It acts as a dial on the entire payout system.

The payout factor started at 1.0 for early rounds, dropped steeply through rounds 250–450, and has stabilized near 0.10 since round 500 — roughly a 10x compression from the early era. That means a model earning 5 NMR per 100 staked in 2019 would earn closer to 0.5 NMR on the same score today. The live curve sits next to earned and burned NMR on the round economics view.

When the payout factor drops, strong scores earn less and bad rounds burn less; when it rises, both sides are amplified. Participants who staked heavily during high-factor periods without strong models learned expensive lessons. A declining payout factor environment means you need better raw scores to achieve the same NMR return.

Distribution of Per-Round Returns

Looking at averages can be misleading. What does the typical staker actually experience on a round-by-round basis?

Per-round returns are tightly packed around zero. The median round returns 0.05% of stake — a rounding error on any single round. Most of the mass sits inside ±2%, with a thin positive tail stretching further out than the negative tail, consistent with the aggregate net-positive number from earlier.

That shape explains why staking can feel unrewarding even in a net-positive system. A 0.05% median round return means many participants go through long flat stretches before compounding shows up. Models that avoid deep burns tend to outperform over time, even if they rarely top the leaderboard — a dynamic explored in Model Survival.

Practical Takeaways

Yes, staking is net profitable in aggregate. Roughly 2.5M NMR earned against 1.1M NMR burned across the tournament's history. That headline masks a median round return of 0.05% and a lot of variance between individual stakers.

What the data suggests:

• Model quality is the main driver. Stake tier does not confer a percentage-return edge — larger tiers just move more absolute NMR because they stake more.
• Watch the payout factor. It has compressed from 1.0 to around 0.10 over the tournament's life. A halved factor means you need double the raw model performance for the same NMR return.
• Consistency beats volatility. The return distribution favors models that avoid large burns. Slightly-positive-most-rounds compounds better than big-swings.
• Size your stake to your conviction. The 25% per-round burn cap means a streak of bad rounds can erode your stake quickly, and burned NMR is gone permanently.
• Time horizon matters. With a median round return near 0.05%, short-term results are mostly noise. Expected value only shows up over dozens of rounds.

Staking on Numerai is not passive income. It is a skill-based game where the edge comes from building better models, and the data shows that edge exists and is rewarded — but not guaranteed.