The VIX Term Structure Manual

Research note. Every figure computed at render from Cboe VIX and VIX3M daily closes, 2010–present — the same file behind the live VIX term structure page. Methods in How we checked it; editorial standards in How we use AI.

What is the VIX term structure?

Two snapshots of the same fear, at two distances: the VIX prices expected S&P volatility over the next 30 days, VIX3M over the next three months. Dividing them gives one number for the shape of the curve. Most of the time the ratio sits below 1.0 — contango — because uncertainty naturally piles up with time; more can go wrong in three months than in one. Today the ratio is 0.85 (VIX 16.5 against VIX3M 19.3), and the curve has been in contango for 82 straight sessions. When the ratio crosses above 1.0 — backwardation — the market is paying more for protection now than for protection later. That inversion is rare, and it is the whole reason to watch this gauge: it has happened on just 8% of days since 2010.

How the number is built — and a lesson we paid for

Both indices come from Cboe, computed from S&P 500 option prices; we track their daily closes back to 2010 and flag every session with the ratio at or above 1.0. One mechanical detail deserves more respect than it gets: Cboe's indexes keep printing after the 4:15pm settlement, so a same-day "close" scraped in the evening can differ from the official settle. We learned this the hard way — in July 2026 we published a backwardation flip off a same-day print that the official settlement erased. Our pipeline now verifies each close against Cboe's own settle the next session, and anything read intraday from this gauge should be treated as provisional until then.

Is backwardation bearish? Mostly the opposite — briefly

Here is the record, which surprises most people. Since 2010 there have been 63 backwardation episodes (runs separated by more than five sessions), with a typical length of about 5 days. From each episode's first day, SPY averaged +3.9% over the next quarter with 75% positive — better than the +3.2% all-days baseline. The average inversion was a panic that passed, and buying it worked more often than not. The danger sits in the tail: the episodes that didn't resolve in a few days were 2011's debt-ceiling summer, late 2018, and the 2020 crash. Duration is the signal. A flip is information; a residence is a regime.

The famous inversions, from our data

And the marathons: the four longest episodes in our record ran 43 days (February 2020), 32 days (July 2011), 17 days (October 2018), 17 days (March 2025) — the 2020 crash, the 2011 debt-ceiling summer, and their relatives. Note what's missing from the deep-print list: nothing since 2020 has come close to a 1.3 reading. The August 2024 yen-carry morning peaked at 1.14 and was gone in days.

Where this gauge will mislead you

• 1.0 is a line, not a cliff. A 0.97 afternoon carries most of the same information as a 1.01 print. Watch the drift toward the line, and the time spent near it, more than the crossing itself.
• Same-day readings are provisional. The settlement quirk above. If a discrete claim depends on today's close, wait for the official settle — we do.
• The ratio ignores the level. Backwardation with the VIX at 25 and at 60 are different animals; this gauge's companion is the VIX level itself — read them together.
• The sample starts in 2010. Our record contains no 2008. Every base rate above describes a mostly rising, post-crisis market, and the buy-the-flip statistic leans on exactly that regime.

How we checked it

Episodes are runs of ratio ≥ 1.0 closes separated by more than five sessions; forward returns are SPY averages from each episode's first day, graded only where the window is complete, and adjacent episodes' windows can overlap. The deep-print table reads each day's ratio from the same dataset the charts draw. Everything recomputes at render, so this Manual always matches the live page — including after Cboe settlement corrections.

Frequently asked questions