Do stacked-up distribution days actually warn of a falling S&P 500? We tested the same rules behind our 20-Day A/D Score against five years of SPY daily data, October 2021 through October 2, 2026 (1,255 sessions). That window covers the 2022 bear market, the 2023–26 bull run and the April 2025 tariff selloff.
The rules
- Distribution day: SPY closes down 0.2% or more on higher volume than the prior session.
- Accumulation day: SPY closes up 1.0% or more on higher volume than the prior session.
- Counts use a rolling 20-session window, the same window as the sidebar score.
The short answer
- Direction: no edge. With 5 or more distribution days in the window, SPY’s average 20-day forward return was +0.6% to +1.6%. A typical day averaged +0.9%. A bootstrap test puts the difference squarely within chance (p = 0.51).
- Risk: a real signal. With 5+ distribution days, the odds of a 5%+ drop within the next 20 sessions were 27–28%, versus 9% at 3–4 days and 19% on an average day.
- Regime matters. The first four clusters of 2022 each led to 7–11% declines within 20 sessions. From 2023 on, most clusters were followed by gains.
[INSERT CHART 1: SPY vs. 20-day distribution-day count]
[INSERT CHART 2: Odds of a 5%+ drop by distribution-day count]
Forward returns by distribution count
| Bucket | Days | Avg 5d | Avg 20d | 20d up | Avg 60d | 5%+ drop |
|---|---|---|---|---|---|---|
| All days (baseline) | 1236 | +0.23% | +0.91% | 63% | +2.78% | 19% |
| D-count 0-2 | 133 | +0.20% | +0.15% | 54% | +3.27% | 14% |
| D-count 3-4 | 479 | +0.08% | +1.07% | 68% | +2.73% | 9% |
| D-count 5-6 | 409 | +0.30% | +0.62% | 59% | +2.30% | 27% |
| D-count 7+ | 215 | +0.44% | +1.62% | 64% | +3.51% | 28% |
Net score (accumulation minus distribution)
| Bucket | Days | Avg 5d | Avg 20d | 20d up | Avg 60d | 5%+ drop |
|---|---|---|---|---|---|---|
| All days (baseline) | 1236 | +0.23% | +0.91% | 63% | +2.78% | 19% |
| Net -4 or lower | 586 | +0.36% | +1.20% | 65% | +2.69% | 25% |
| Net -3 to -1 | 507 | +0.06% | +0.69% | 61% | +2.39% | 13% |
| Net 0 to +1 | 112 | +0.35% | +0.78% | 65% | +5.02% | 13% |
| Net +2 or higher | 31 | +0.16% | -0.36% | 48% | +2.56% | 29% |
The correlation between the 20-day count and the next 20-day return was +0.07 for distribution days and −0.09 for the net score. Both are effectively zero.
Every cluster signal
A signal is the first day the count reached 5 after being below it, with at least 20 sessions between signals. “Max drop” is the worst close in the following 20 sessions.
| Signal date | SPY close | Next 20d | Next 60d | Max drop |
|---|---|---|---|---|
| 2021-11-30 | 455.56 | +4.8% | -3.9% | -1.1% |
| 2022-01-13 | 464.53 | -5.2% | -5.3% | -7.2% |
| 2022-04-06 | 446.52 | -7.3% | -14.5% | -7.7% |
| 2022-06-08 | 411.22 | -5.5% | -4.6% | -11.0% |
| 2022-08-30 | 398.21 | -7.0% | +1.1% | -8.7% |
| 2022-11-02 | 374.87 | +8.7% | +8.4% | -1.0% |
| 2022-12-15 | 389.63 | +2.1% | -0.1% | -3.3% |
| 2023-03-01 | 394.74 | +1.7% | +5.0% | -2.4% |
| 2023-05-02 | 410.84 | +1.7% | +11.2% | -1.4% |
| 2023-06-07 | 426.55 | +2.8% | +5.8% | +0.6% |
| 2023-08-08 | 448.75 | -0.6% | -5.8% | -2.8% |
| 2023-09-19 | 442.71 | -1.5% | +6.3% | -4.8% |
| 2024-03-08 | 511.72 | +1.4% | +3.3% | -0.4% |
| 2024-04-15 | 504.45 | +3.3% | +10.3% | -1.8% |
| 2024-06-20 | 547.00 | +0.4% | +2.9% | -0.8% |
| 2024-07-24 | 541.23 | +3.6% | +7.6% | -4.4% |
| 2024-08-28 | 558.30 | +2.5% | +6.3% | -3.2% |
| 2024-10-31 | 568.64 | +6.0% | +6.4% | +0.2% |
| 2024-12-27 | 595.01 | +1.1% | -4.7% | -2.4% |
| 2025-02-03 | 597.77 | -3.5% | -7.2% | -3.5% |
| 2025-06-17 | 597.53 | +5.1% | +10.0% | -0.5% |
| 2025-07-31 | 632.08 | +2.7% | +7.1% | -1.6% |
| 2025-09-02 | 640.27 | +4.0% | +5.4% | +0.5% |
| 2025-10-10 | 653.02 | +2.7% | +5.6% | +1.2% |
| 2025-11-18 | 660.08 | +1.7% | +3.4% | -1.1% |
| 2025-12-29 | 687.85 | +1.1% | -6.2% | -1.5% |
| 2026-02-26 | 689.30 | -6.4% | +8.2% | -6.4% |
| 2026-05-07 | 731.58 | +0.8% | +5.4% | +0.3% |
| 2026-06-09 | 737.05 | +2.0% | +4.9% | -1.6% |
| 2026-09-01 | 761.78 | +0.1% | n/a | -1.0% |
Of 29 completed signals, 8 (28%) saw SPY lower 20 sessions later, compared with 37% of all 20-day windows. A 5%+ drop followed 5 of them: four in 2022 and one in February 2026.
A stricter definition
Our rules flag about 1 in 4 sessions as a distribution day, so the 20-day count averages around 5. As a check, we re-ran the test counting only days down 1% or more on volume above the 50-day average. The pattern held. Five or more of those days lifted the 5%+ drop odds to 29%, versus 13% when there were none. But the average 20-day return was higher (+2.1% versus +0.2%), because heavy selling tends to cluster near short-term lows.
What we take from it
A high distribution count is a risk gauge, not a sell signal. It says the tape is volatile and a sharp drop is more likely than usual. It does not say the market will be lower a month from now. In a downtrend like 2022 it was an excellent warning; in a bull market it mostly flagged shakeouts that were bought.
Method and limits
- Data: SPY daily bars from Interactive Brokers, which caps daily history at five years. A longer test would cover more market cycles.
- Prices are not dividend-adjusted, so a few ex-dividend days may register as borderline distribution days.
- Forward windows overlap, so daily samples are not independent; the cluster table and the bootstrap test account for this.
- This is a historical study, not investment advice.