COST 20-day forward returns: near-ATH (≤2% below) vs 10%+ drawdown — last ~3 years
The short answer is no: over the last ~3 years, buying COST within ~2% of its trailing all-time high did not deliver superior 20-day returns versus buying when the stock was in a 10%+ drawdown. The two buckets produced similar short-term outcomes — mean 20-day returns of 1.78% (near-ATH, n=247) versus 2.22% (10%+ drawdown, n=150), a −0.44 percentage-point difference that the sample doesn’t support as reliable.
Below you'll find the full setup and statistics: we flagged days by an expanding ATH, measured close[t+20]/close[t] − 1, and compared distributions (means, medians, hit rates) with a Welch t-test (t = −0.75, p = 0.456). Medians and hit rates tell a mixed story and overlapping 20-day windows reduce effective power — the detailed charts and tests are in the report that follows.
For COST over the past ~3 years, did buying when it closed within ~2% of a trailing all-time high deliver better forward 20-day returns than buying when it sat in a 10%+ drawdown — i.e.
How this was measured
Resampled COST minute bars to daily closes and limited the window to the last ~36 months. Computed an in-sample trailing all-time high via expanding max of close; drawdown = close/ATH − 1. Flagged days within 2% of ATH (drawdown ≥ −0.02) and days in 10%+ drawdown (drawdown ≤ −0.10). For each flagged day, measured the 20-trading-day forward return as close[t+20]/close[t] − 1. Compared bucket distributions (means, medians, fraction positive) and applied a Welch t-test on mean differences.
The key numbers
Reading the numbers
Across 1,094 days, there were 247 near‑ATH (≤2% below) events with a mean 20‑day return of 1.78% and 150 deep drawdown (≥10%) events with mean 2.22%. The drawdown edge is only ~0.44 percentage points and the p‑value (0.456) says that difference is not statistically clear.
The charts
The box plot lays out the full return distributions: Near‑ATH (n=247) has mean 0.0178 and spans -0.1632 to 0.1532; 10%+ drawdown (n=150) has mean 0.0222 and spans -0.0709 to 0.1497. Look at the much deeper negative tail for the Near‑ATH bucket (worst -16.32%) versus the shallower worst loss in the drawdown bucket (-7.09%): that asymmetry helps explain why means are close even though the Near‑ATH bucket had a higher fraction of positive outcomes.
The bar chart highlights the simple means: 1.78% for Near‑ATH and 2.22% for 10%+ drawdown. The difference is small (mean difference Near minus Deep = -0.0044277, i.e., about -0.44 percentage points) — visually the bars are close and the gap is not large enough to be decisive on its own.
The Near‑ATH histogram shows 247 observations with returns ranging from -16.32% up to +15.32% and a mean of 1.78%. What to watch is the presence of sizable negative outliers down to -16.32%: although 61.54% of these trades were positive, those rare big drops pull the distribution leftward.
The 10%+ drawdown histogram has 150 observations, ranging -7.09% to +14.97% with a mean of 2.22%. Compared with the Near‑ATH histogram, this bucket shows a shallower downside (worst -7.09%) and a slightly higher average despite a slightly lower positive fraction (56%), which is why the drawdown mean edges the Near‑ATH mean in the summary stats.
Bucket summary — 20-day forward returns
| bucket | N | mean | median | std | fraction_positive |
|---|---|---|---|---|---|
| Near ATH (≤2%) | 247 | 0.0178 | 0.0209 | 0.0598 | 0.6154 |
| 10%+ drawdown | 150 | 0.0222 | 0.0077 | 0.0557 | 0.56 |
The takeaway
No — buying within ~2% of a trailing ATH did not produce better 20-day returns than buying in a 10%+ drawdown over the last ~3 years. There were 247 near-ATH days versus 150 deep-drawdown days; the mean 20-day return was 1.78% for near-ATH and 2.22% for drawdown, a -0.44 percentage-point difference. Medians and hit rates tell a mixed story: median return was higher for near-ATH (2.09% vs 0.77%) while the fraction of positive 20-day returns was 61.5% vs 56.0%. Statistical testing shows this gap is indistinguishable from noise: Welch t = -0.75 with p = 0.456, and overlapping 20-day windows further reduce effective sample power. Bottom line: there’s no reliable short-term edge here — the differences are small and inconclusive for COST in this period.
The fine print
- The trailing ATH was computed only inside the ~3-year window, so early-period ATH proximity can be misclassified.
- Forward 20-day windows overlap heavily, which reduces effective sample size and inflates apparent confidence.
- Returns are close-to-close, price-only (no dividends) so total-return numbers differ slightly from these figures.
- Signals use closing prices only; intraday new highs that fade by the close are not counted as near-ATH.