PEP performance on SPY's worst-decile down days — defensive or not?
Conventional wisdom says PepsiCo is a safe harbor when markets turn ugly. But over the past three years, on the S&P 500's worst 10% of days—when the index averaged a 1.66% loss—PepsiCo’s average return was just -0.04%. That minuscule drawdown is statistically indistinguishable from zero, and the 1.62 percentage point cushion over SPY is rock-solid, with odds of being random luck near zero.
This analysis tests the defensive claim head-on. Using daily close-to-close returns from the last ~750 trading days, we isolated the 76 crash days for SPY and measured how PepsiCo behaved. The full breakdown—means, medians, fraction positive, and the paired t-test—is below. The verdict: the safe-haven reputation holds up on the days it matters most.
Over the past ~3 years, does PepsiCo actually hold up as a defensive stock on the market's worst days, or does it fall nearly as hard as SPY on its worst-decile down days? Thesis: PEP's average return on SPY's crash days is deeply negative, so the safe-haven reputation fails when it's needed most.
How this was measured
Resampled SPY and PEP minute bars to daily close-to-close returns. Defined SPY's worst-decile days as those with daily return ≤ 10th percentile. On those crash days, computed PEP's mean, median, fraction positive, and the paired difference (PEP minus SPY). Tested whether this difference is zero (paired t-test) — a statistically significant positive difference would support defensive behavior; a result indistinguishable from zero means PEP falls roughly with the broad market when stress hits. Also compared PEP's crash-day mean to its unconditional long-run mean. Data window: last 3 years, ~750 trading days, ~75 crash days.
The key numbers
Reading the numbers
On SPY's worst 10% days (average loss -1.66%), PepsiCo's average return was just -0.04%—barely a blip. 42% of the time PEP actually closed up on those crash days. The safe-haven reputation holds.
The charts
This box plot compares daily returns on SPY's worst-decile days. SPY's entire box sits below -0.97%—every crash day is a loss. PEP's box straddles zero, with a mean of -0.04% and many positive outliers above +2%. The difference in central tendency is massive: PEP barely budges while SPY falls 1.66% on average.
This histogram shows all 76 PEP returns on SPY crash days. The bars cluster around zero, with roughly equal counts on either side. The mean is -0.04%, but the range is wide: from -3.6% to +5.5%. Contrast this with a typical defensive stock that might fall in sympathy—PEP's distribution is centered near zero, not deep negative.
This bar chart compares three mean returns. PEP's unconditional mean over 751 days is -0.02%, nearly flat. On crash days it's -0.04%, essentially unchanged. SPY's crash-day mean is -1.66%, a huge negative bar. The thesis that PEP falls hard on bad days is refuted: its crash-day average is indistinguishable from its long-run average, while SPY plunges.
First 20 worst-decile SPY days (most recent in range)
| date | spy_return | pep_return | spy-pep_gap |
|---|---|---|---|
| 2023-08-15 | -0.0118 | -0.0113 | -0.0006 |
| 2023-08-24 | -0.0178 | -0.0001 | -0.0177 |
| 2023-09-15 | -0.0129 | -0.006 | -0.0069 |
| 2023-09-20 | -0.0113 | -0.0001 | -0.0112 |
| 2023-09-21 | -0.0155 | -0.0174 | 0.0019 |
| 2023-09-26 | -0.0128 | -0.0016 | -0.0112 |
| 2023-10-03 | -0.0137 | -0.0071 | -0.0065 |
| 2023-10-18 | -0.0107 | 0.0139 | -0.0245 |
| 2023-10-19 | -0.0107 | -0.0119 | 0.0011 |
| 2023-10-20 | -0.0122 | -0.0019 | -0.0103 |
| 2023-10-25 | -0.0178 | 0.001 | -0.0188 |
| 2023-10-27 | -0.0107 | -0.0142 | 0.0035 |
| 2023-12-20 | -0.0126 | -0.0128 | 0.0003 |
| 2024-01-31 | -0.0106 | -0.0106 | -0.0001 |
| 2024-02-13 | -0.0119 | -0.0102 | -0.0016 |
| 2024-04-04 | -0.0139 | -0.0009 | -0.0129 |
| 2024-04-10 | -0.0121 | -0.0149 | 0.0028 |
| 2024-04-12 | -0.016 | -0.002 | -0.0139 |
| 2024-04-15 | -0.0108 | -0.0077 | -0.0031 |
| 2024-04-30 | -0.0167 | 0 | -0.0167 |
The takeaway
Over the past ~3 years, on SPY's worst 10% of days, PepsiCo actually does hold up as a defensive stock — its average loss is barely a blip compared to the broad market. On those 76 crash days, PEP fell just 0.04% on average, while SPY dropped 1.66%. That 1.62% edge is statistically rock-solid: the p-value is essentially zero, so there's virtually no chance this cushion is random luck. PEP still had a positive return on 42% of those crash days, only slightly below its normal 49% win rate, and its crash-day average of -0.04% is very close to its overall daily average of -0.02%. So the safe-haven reputation looks justified: during SPY's worst selling days, PepsiCo does not fall nearly as hard. The thesis that PEP's crash-day average is deeply negative is not supported by this data — it barely loses money at all, making it a meaningful refuge in rough markets.
The fine print
- Worst-decile definition uses in-sample quantile — the threshold is data-driven and can shift over time outside this period.
- Crash days are not independent; volatility clustering may inflate the effective sample size, and the t-test assumes roughly normal differences despite fat tails in daily returns.
- Results cover only the last ~3 years; a different window (e.g., including more severe bear markets) could produce a different threshold and possibly larger or smaller defensive behavior.