Attention fades. A coin that spikes alone lags Bitcoin the following week.

Alone, the lag is clear. In company, it is not.

Breadth is the share of scored coins at 70 or more on the same day. A lone spike is one of under 3%. When many coins spike together the whole market is moving, and one coin's spike says little about that coin. Across all 2,461 upside spikes the median lagged by 3.4%, 1.3 points worse than an ordinary day (-2.1 pts to -0.5 pts). Most of that comes from the lone ones.

In between, the median was -0.1 pts against an ordinary day, not separable from one. Crowd days are withheld: 133 spikes on only 5 dates, 41% of them on November 6 2024. That is too few dates to say anything.

Why the average lies

The average after an upside spike is +0.7%. The median is -3.4%. That gap is the whole story. Returns after a spike are lopsided: most coins give back ground, and a few run far enough to pull the average up. The runners are the ones people post about. The median is what the typical buyer got.

The biggest rises stand furthest apart

The median lag grows with the size of the spike-day rise, but once whole dates are resampled only some bands separate from an ordinary day. A rise of 40% or more is the clearest: -13.9% over the next week, 128 cases.

Year by year

Lone spikes finished below an ordinary day in every one of the 5 years. All upside spikes together did in 4 of 5. 2026 runs to September 21 2026.

Downside spikes: nothing to report

752 downside spikes on 194 dates. The median lagged Bitcoin by 1.3% over the next week, +0.8 pts against an ordinary day, with an interval of -1.4 pts to +3.4 pts. That is not distinguishable from an ordinary day. One date, October 10 2025, holds 102 of them. Without it the median is -2.5%, beat 38% of the time, and still not distinguishable.

Stocks: not measured yet

The Terminal also scores 43 US stocks and ETFs besides SPY. They have no base rate. The same study, measured against SPY instead of Bitcoin, is written: the live formula on daily bars since 2018, the same threshold of 70, the same breadth and withholding rules. It needs daily prices and volume back to 2018 from a source that allows research use without a key. On September 22 2026 none qualified: the free sources sit behind a browser check, require a key, or limit use to personal purposes. Until it runs, the API and the MCP tool return no base rate for stocks and say why, and asset pages show none.

For stocks, a scheduled results release is a common reason for a day of unusual volume, so each spike will be labeled by whether it fell within a day of one. The earnings calendar is in hand: 1,282 results filings (SEC form 8-K, Item 2.02) since 2018, for 39 of the stocks. The ETFs file none, and ARM and SPOT report on form 6-K, so spikes in ARM, QQQ, SPOT and USO will carry no earnings label.

Method, and what this is not

• Data: Binance public spot market data, daily, USDT pairs, 2022-01-01 to 2026-09-21, 175 coins. Stablecoins, wrapped tokens, tokenized stocks and gold excluded. Bitcoin is the benchmark.
• A spike is a day the live Activity score formula, run on daily candles, reached 70, the same threshold that opens an event on the Terminal. The same two inputs, 24h volume and the absolute move, measured on Binance daily candles rather than the live feed. The same 30 day baseline with a 14 day minimum, the same clamp, the same weights. One spike per coin per week, so overlapping windows are not double counted.
• Every outcome is the coin's return minus Bitcoin's over the identical window, starting at the close of the spike day. Ordinary days are a random sample of 12,945 days on the same coins where the score stayed well below the threshold.
• 5 spikes are left out because the coin stopped trading inside the window, from a delisting or a data gap. They are not counted as losses, which flatters the spike results slightly.
• Many coins spike on the same day when the whole market moves, so cases are not independent. Every interval resamples whole dates. A group is withheld when it has fewer than 100 spikes or one date holds more than 20% of them, and marked not separable when its interval against an ordinary day includes zero. The same rule applies on asset pages, the API and the MCP tool.
• This is a historical measurement. Fees, slippage and the practical difficulty of shorting small coins are not modeled. It is not advice to buy or sell anything.

Download all 3,323 spikes as JSON · The live record, since August 2026

One row per spike: symbol, date, day_move_pct, strength (the Activity score that day), breadth_share, and excess_1d, excess_7d, excess_30d against BTC in percent. 105 recent spikes have windows that have not finished, and 5 ended early with a delisting or a data gap. Both carry null.

Corrections

September 22 2026

• The first version of this page ran a different formula from the live score: log volume, equal weights and a lower cutoff. It now runs the live formula, and every figure here was recomputed.
• Withdrawn: “attention on the way down tends to be recovered.” The largest downside group, a 20 to 40% drop, had 186 cases and beat Bitcoin 64% of the time, but 102 of those cases fell on one date, October 10 2025, a day most of the market fell at once. Without that date the group lagged Bitcoin by a median of 4.5% and beat it 38% of the time. Under the live formula downside spikes show no difference from an ordinary day, with or without that date.
• Intervals now resample whole dates rather than single spikes, and groups that are small, dominated by one date, or not distinguishable from an ordinary day are withheld. The base rates on asset pages, the API and the MCP tool follow the same rule, and are now matched on breadth.