Stake Flows: When Smart Money Moves

Net NMR outflows lead burn events by 1-2 rounds, large stakers (1K+ NMR) cut positions ~3% faster than small stakers after burns, and 62% of all stakes sit unchanged round to round.

Stake is not static. Each week, participants add NMR, withdraw it, or leave their position untouched. Those flows carry signal. When net stake turns negative ahead of a rough stretch, someone saw it coming.

This post measures stake flow patterns in the Numerai tournament: whether outflows lead or lag burn events, how stake tiers respond differently, whether NMR volume tracks tournament difficulty, and how passive most stakers really are. For background on staking mechanics, see Is Staking Profitable?. Live data sits on the trends page.

Net Flow vs. Burn Rate

The first question is directional: does capital leave the tournament before burns, or after?

Dual-axis line chart of net NMR stake flow per round overlaid with rolling burn rate from rounds 200 through 1200, showing net outflows consistently leading burn spikes by 1-2 rounds
Dual-axis line chart of net NMR stake flow per round overlaid with rolling burn rate from rounds 200 through 1200, showing net outflows consistently leading burn spikes by 1-2 rounds

Net flow — the change in total staked NMR — swings between +50K and -80K NMR in extreme rounds. The rolling burn rate (share of staked models with negative payout, 10-round average) tracks a similar rhythm but lags visibly. Outflows precede burn spikes by 1-2 rounds rather than follow them.

That lead suggests a subset of participants — likely those with better models or more experience — adjust on early score signals before final resolution. When net flow turns sharply negative after stability, the next few rounds deserve attention. See Market Regimes for how burn events cluster.

Behavior by Stake Tier

Not all participants respond to stress the same way. Breaking models into stake tiers reveals how size shapes behavior.

Grouped line chart showing average stake change percentage per round for four tiers: sub-10, 10-100, 100-1K, and 1K+ NMR, with larger tiers showing faster drawdowns after burn events
Grouped line chart showing average stake change percentage per round for four tiers: sub-10, 10-100, 100-1K, and 1K+ NMR, with larger tiers showing faster drawdowns after burn events

Large stakers (1K+ NMR) reduce their positions roughly 3% faster than sub-10 NMR stakers in the round following a burn event. The 100-1K tier sits in between, and sub-10 stakers are the most inert — small stakes hurt less, and smaller participants tend to be more passive.

The gap narrows during calm periods. When burns are low, all tiers drift upward at similar rates. The divergence opens specifically during stress, consistent with larger stakers running active risk management. This tiered response partly explains the leading outflow signal above: the largest positions move first and move the aggregate number.

For a live breakdown of how stake concentrates among top models, see the stake concentration chart.

NMR Volume vs. Tournament Stress

If tournament burns create sell pressure, that should show up in exchange trading volume.

Dual-axis chart of NMR daily trading volume alongside per-round burn rate, with volume spikes loosely coinciding with high-burn periods but also driven by broader crypto events
Dual-axis chart of NMR daily trading volume alongside per-round burn rate, with volume spikes loosely coinciding with high-burn periods but also driven by broader crypto events

The relationship is noisy. NMR volume rises during high-burn stretches, but also spikes on crypto-wide events unrelated to the tournament. The correlation is positive but modest — enough to confirm tournament stress generates some sell pressure, not enough to isolate it from the broader market.

The cleanest signal is in the tails: the five highest-volume weeks all coincide with either a burn spike above 65% or a major crypto move, usually both. For the supply-side view, see NMR Token Economics.

Stake Stickiness

Most stakers do not actively manage their position. "Stake stickiness" measures the share of models with identical selected_stake_value to the prior round.

Line chart of stake stickiness percentage over time, hovering around 62% with dips to 45% during high-volatility periods and peaks near 75% in calm stretches
Line chart of stake stickiness percentage over time, hovering around 62% with dips to 45% during high-volatility periods and peaks near 75% in calm stretches

On average, 62% of models carry the exact same stake from one round to the next. In calm stretches stickiness climbs to 75%; during volatile periods it dips to 45%, meaning a majority are adjusting. Those dips align with the stress periods visible in every chart above.

High stickiness stabilizes total stake and makes the meta-model more predictable. But the active minority — the 25-38% who adjust — exerts outsized influence on aggregate flow. When stickiness drops and net flow goes negative simultaneously, that convergence has historically preceded the worst burn clusters.

Takeaways

Net outflows lead burn events by 1-2 rounds. Capital begins leaving before final round resolution, suggesting informed participants adjust on early signals rather than waiting for results.

Large stakers respond faster to stress. The 1K+ tier cuts positions ~3% more aggressively than smaller tiers after burn events. In calm periods, tier behavior converges.

NMR volume loosely tracks tournament stress. High-burn periods coincide with elevated trading volume, but crypto-wide moves dominate the signal. Volume alone is not a reliable tournament indicator.

Most stakes are passive. 62% of positions go unchanged round to round. The active minority drives aggregate flows, and drops in stickiness combined with negative net flow have historically preceded difficult stretches.