Market structure · Aug 2026
There’s a comfortable idea that bull and bear markets alternate on a schedule — similar time between similar events, a rhythm you could lean on. I tested it on nearly a century of S&P 500 data, then kept pulling the thread until it reached a question worth money: if there’s structure here, can you trade it?
Every peak looks obvious in hindsight, and hindsight is where the “cycle” story lives. So here is the honest version. I took the S&P 500 price index back to January 1928 — nearly 98 years, the longest continuous series available — and dated every bull and bear market with the standard 20% rule: a bear begins when the index falls 20% from a peak, a bull when it rises 20% from a trough, each turning point dated at the actual extreme. That yields 27 completed bull markets and 27 bears. Then I asked three questions of them in turn: is the timing regular, is the size regular, and if either is — can a signal you’d actually have in real time act on it?
Time between peaks
A 38× spread. If markets ran on a clock, this number would barely move. There is no cycle.
Non-recession bear ceiling
Crashes without a recession cluster below a third. Deeper than that needs a recession.
Best real-time filter
Not the yield curve. Just the market’s own price trend — reacting beats forecasting.
Draw every market to scale on the calendar and the story falls apart on sight. Bull markets range from six weeks to twelve years; the gap between successive peaks runs from four months to 12.6 years. A perfectly regular clock has a coefficient of variation of 0; memoryless random timing sits near 1. Bull-market timing comes in at 1.21 — essentially as irregular as a coin.
And a cycle needs memory — one interval predicting the next. There is none: the lag-1 autocorrelation of bull durations is +0.04, of peak-to-peak spacing +0.16. Both are indistinguishable from zero. The one stable feature is shape, not schedule: bears are short and sharp, bulls long and grinding. That asymmetry is why “time in the market” works — but it is a distribution, not a calendar.
each block is one market on the calendar · width = real elapsed time
Timing is near-random. Does the size of moves have more structure? Partly — and the two sides split in opposite directions. Bear depths are the most regular thing in the dataset: mean −35%, median −33%, and the lowest dispersion of anything measured. Two-thirds of all bears land between −25% and −40%. If you had to name a typical market event, the market falls about a third is the closest a century comes to a constant.
Bull gains are the opposite — a fat right tail. Most are modest (median +79%), but a handful of secular runs dominate: 1949 (+266%), 1982 (+229%), 1987→2000 (+582%), 2009→2020 (+400%). A few bulls carry the whole century’s return, which is the strongest argument there is for staying invested. Two things people expect here and don’t get: no magnitude memory (autocorrelation +0.00 for bulls, +0.13 for bears), and no “deep crash → big rebound” — the correlation between a bear’s depth and the next bull’s gain is −0.20, faintly the wrong sign.
every completed market since 1928 · one dot per market
That tight bear cluster is actually two populations. Split each bear by whether its decline overlapped an official NBER recession and a gap opens up: non-recession bears average −29% with astonishingly low spread; recession bears average −39% and run far deeper. The difference is significant (Welch t-test p = 0.006). The headline is the ceiling near −34%: crashes, panics, and growth scares without a recession — 1987, the 1962 Kennedy Slide, 1966, 2022 — cluster below roughly a third. Every decline meaningfully deeper came with a recession: 1937, 1973–74, 2000–02, 2008, and the Depression legs down to −62%.
each bear by decline depth · Welch t-test p = 0.006
If deep bears are recession bears, maybe a recession signal could sidestep them. I tested this honestly: a monthly, no-lookahead backtest on total-return S&P 500 (1988–2026), cash earning the 3-month T-bill, every signal lagged for real release delays. The benchmark to beat isn’t just buy-and-hold — it’s a plain price trend filter (hold above the 10-month average, else cash), the cheapest way to react to a developing bear.
The textbook leading indicator, the 10y–3m yield curve, inverts before every deep recession bear — but 11 to 27 months early, and it re-steepens before the crash as the Fed cuts, so “spread > 0” flips back to a buy signal exactly when the recession hits. As a standalone rule it matched buy-and-hold’s full −51% drawdown and a worse risk-adjusted return. A warning two years early that un-warns at the worst moment is not a trade.
peak-to-trough of each strategy’s equity · darker = deeper loss
Weekly initial jobless claims and the unemployment-based Sahm rule are coincident — they turn as the recession actually begins, and that timing is the whole game. Both roughly halved the max drawdown where the curve did nothing. The twist: the price trend filter still edges every macro signal and cut the deep bears hardest (−6% in 2000, −4% in 2008). Labor data is reacting to the same recession that breaks the price trend — and price breaks a step earlier, with less give-back. Stacking a macro signal on top of the trend filter adds nothing; they’re substitutes, not complements.
| Rule | CAGR | Max DD | MAR |
|---|---|---|---|
| Buy & hold | 11.5% | −50.9% | 0.22 |
| Yield curve (spread > 0) | 10.1% | −50.9% | 0.20 |
| Initial claims (3m vs 12m MA) | 9.9% | −23.9% | 0.41 |
| Sahm rule | 11.0% | −23.9% | 0.46 |
| Price trend (10-mo MA) | 10.2% | −21.4% | 0.47 |
The yield curve is right in direction and hopelessly early. Labor data is coincident and halves the drawdown. But the best real-time read on a developing recession bear is still the market’s own price.
Macro data confirms the bear — price times it.$SPX) daily, Jan 1928–Aug 2026. Backtests use the total-return index ($SPXTR) from 1988.%10YTCM/%3MTCM; claims #INJOB and unemployment #UNEMP from Norgate’s Economic database — no separate FRED pull needed. All lagged one month, no lookahead.code/ folder.The unifying lesson is react, don’t forecast. The “every N years” cycle is hindsight pattern-matching; the real regularity is that the business cycle, not the calendar, decides how deep a bear goes. And the information that a recession is turning a selloff into a rout shows up in the price before it shows up cleanly in the macro releases. The number you get from a story about cycles is not the number you’ll trade.