Gaemi score
Does it work?
The score, replayed day by day over history using only what was public at the time, then checked against what each stock did next.
What we found
- Volatility: after a score of 85+, the stock moved 1.08× its usual daily amount over the next week, against 1.03× at scores under 30.
- Direction: 20 trading days after a score of 85+, stocks were +0.2% vs SPY on average (±2.0). That's within the noise: the score doesn't reliably say which way the price goes.
- Ranking stocks by score each day: correlation 0.021 with next week's volatility, 0.008 with next week's return. The volatility link is clear of chance; the return link isn't.
The test
- Period
- 2015-10-09 → 2026-10-02
- Stock-days
- 278,230
- Stocks
- 130
- Compared with
- SPY
Data: Bluesky posts, GDELT world news, Hacker News, Finnhub headlines, Wikipedia page views; news from 2026-07-07.
Return next 5 days
vs market, by score
Return next 20 days
vs market, by score
Volatility next week
× usual, by score
Whiskers are 95% intervals; the dotted line is the average over all stock-days; small numbers under each bar are how many stock-days fell in that score range.
After a spike
score jumps to 70+ after 10 quiet days
━ all spikes (2419), shaded 95%━ no news (4)
Rank correlation
score vs what came next, per day
| Outcome | Average | 95% | Days |
|---|---|---|---|
| Return next day | 0.006 | ±0.004 | 2,760 |
| Return next 5 days | 0.008 | ±0.010 | 2,756 |
| Return next 20 days | 0.002 | ±0.020 | 2,741 |
| Volatility next week | 0.021 | ±0.009 | 2,756 |
Each day, stocks are ranked by score and by the outcome; 1 would mean the order always matched, 0 no link. Highlighted = clear of chance at 95%.
How to read this
- Every score is recomputed as it would have been that day, from data public by its close: no peeking ahead.
- Returns are compared with SPY, so a market-wide rally doesn't count as a signal.
- Volatility is the average daily move over the next 5 days divided by the average over the 60 days before.
- Uses today's list of tickers, so companies that failed or were delisted are missing (survivorship).
- Before each opinion source existed, scores rest on attention (Wikipedia) and news alone.
- Returns are log returns over the market ticker, before trading costs and taxes.
- Past patterns can fail in the future. This is research, not investment advice.