Does the Hindenburg Omen work?

Give the Omen its due first, because the core idea is sound. A healthy advance is broad: lots of stocks making new highs, almost none making new lows. When you see both wings firing at once — a pile of new highs and a pile of new lows, while the index itself is still grinding upward — something underneath the surface is coming apart. That’s a real form of breadth divergence, and most people never look past the index price to see it. Reaching for that signal is a good instinct.

The trouble is the reputation. The Hindenburg Omen gets invoked as a crash siren — fire it and batten the hatches. So I coded the classic rules (both wings above 2.2% of issues, index above its level 50 days ago, McClellan Oscillator negative, new highs no more than twice new lows), ran it across the S&P 1500 back to 1995, and measured what SPY actually did afterward. Two things fell out. There is a real edge in there — it’s just small and short. And the crash-prediction part barely beats a coin the market was already flipping.

01What actually shows up: a one-month chill

Here’s the honest edge. Line up all 214 signal days and track SPY forward, and the first month runs slightly negative where an average month runs positive. Five days out: −0.16% versus a +0.24% baseline. Ten days: −0.17% versus +0.48%. Twenty days: −0.68% versus +0.95%. Just over half of those windows finish lower, against roughly a third for a random day. That’s a genuine, repeatable bearish tilt — the market really does trade a touch heavier right after the internals split.

But watch what happens next. Push the horizon out and the chill wears off: by 40 days the average is back above zero, and by 60 the drift is positive again. The signal buys you a cautious month, then the market’s upward pull reasserts itself. It’s a speed bump, not a wall.

02The crash bell barely rings

Now the part everyone actually cares about: does a confirmed Omen warn of a real drop? I took every 2+ and 3+ cluster and asked whether a meaningful drawdown followed within 60 days — then compared it to the rate on a random day, because the market takes ordinary dips all the time. That baseline is the whole trick, and it’s where the Omen’s reputation quietly falls apart.

A >10% drawdown inside 60 days follows a 2+ cluster 20% of the time — exactly the baseline rate. The stronger 3+ cluster does nudge ahead: 30% versus 20% for a 10% drop, and 19% versus 9% for a 15% drop. So the intense clusters do carry a bit of extra crash risk. But “19%” means that even the scary version is followed by no big drawdown 81% of the time. As a bell that tells you to sell, it rings false far more often than true.

03In fairness: it did fire before the big ones

The false-alarm rate is the headline, but it would be unfair to stop there — because when the market genuinely broke, the Omen was often standing there first. It flagged October 2007 on the doorstep of the financial crisis, May 2008 before the worst of it, late January 2020 three weeks ahead of the COVID crash, April 2022 into that bear market, and February 2025 before a sharp 19% air pocket. Those aren’t nothing. The signal really is drawn to unhealthy tape.

The problem is everything else it flagged. The same rule cried wolf in 2004, 2017, 2019, 2023 and twice in 2024 — clusters that were followed by the market simply going up. Below is a fair sample of both faces: the hits that made its name, and the misfires that make it untradeable on its own.

04So what do you actually do with it?

Treat it as a veto, not a forecast — and keep the window short, because that’s where the entire edge lives. If you simply step to cash for 20 trading days after every signal and otherwise buy and hold SPY, you come out ahead on every axis: 12.5% a year versus 11.2% for buy-and-hold, a shallower worst drawdown (−49.8% versus −55.2%), and a better Sharpe (0.78 versus 0.66) — all while sitting in cash only about a quarter of the time. You’re just monetizing that one-month chill from section 1.

Push past 20 days, though, and it backfires: holding out 40 or 60 days sits out the recovery and drags the return below buy-and-hold. Waiting for the rarer 2+/3+ clusters doesn’t help either — too few events to move a portfolio, and the edge is diluted by then. The usable version is unglamorous: a light, brief risk-off nudge after any signal. Not an alarm to liquidate on.

None of this makes the underlying idea wrong — and that’s the fun of it. Breadth divergence is a legitimate thing to watch, and the Omen was an honest attempt to formalize it. Chasing it one rabbit hole deeper just right-sizes the claim: it’s a mild, one-month caution flag with a real but modest edge, not the apocalypse siren the name implies. Respect it as a small yellow light. Don’t sell your portfolio on it.