Calibrated on the first half of the century. Checked on the second.
The Market Stress Index was fitted on the 12,386 sessions between 1927 and 1977, and then read the 12,386 that followed without being re-tuned once. It measures the state of the market, not what comes next.
S&P 500 · historical replay, not a live reading
Elevated
1929
Historical median 58 / 100 · states at their real share of the scale
MSI works as a precision thermometer: it reads the structural tension in the market from its price history, and from the S&P 500 only.
Most risk measures describe how much prices have been moving. This one describes the structural state of the series itself.
The result is a single reading from 0 to 100: normal, elevated, high or severe.
The method is stated rather than hidden, and the same price history always produces the same reading. Nothing in the output is a black box.
The same reading across the whole history. No retraining, no per-period tuning, no regime-switching.
It reports a state to interpret, not an alert to act on. It does not tell you what to buy, sell or hold.
Almost a hundred years of the S&P 500, read by the index. The decline phases marked here were located from price alone, with one mechanical rule: a retreat of at least 20% from the previous high. The index does not know they exist; what this chart shows is what it was reading while they ran.
The monthly reading of the index, 1929-2026. Shaded: the century's twelve declines of 20% or more, located from price alone.
Ctrl + wheel to zoom · drag to move through time · double click for the full century
The six declines with a name of their own, out of the twelve marked above. The fall is the price. The reading is what the index showed while that fall was running, not a warning issued before it.
Read the full finding for the price on its own, the two series together, and the state of every month of the century, with the method and the limits around them.
The reading was fitted on the first half of the century, and then checked on the half that followed without being re-tuned once. Each figure below says which stretch it is measured on. The decline phases were located from price alone, independently of the index.
The figure that carries the argument is the first one: on the half of the history that took no part in any decision, the index reads 22.5 points higher during decline phases than outside them. It separates the two better there than on the stretch it was fitted on, which is the opposite of what an overfitted model does.
The counts cover the whole century, calibration included, and they are counts rather than a rate. Of the twelve major declines, nine ran with average tension above the historical median. In eight of the twelve, that tension had sat above the median for months beforehand, the same yardstick as the count before it. That describes the state that preceded them; it is not a claim that the index announced anything.
High tension is not a coming decline: episodes that corrected without one lasted 95 sessions at the median; those that ended in a decline, 273.
How long tension lasts
11 to 26 months
Half of the 17 episodes that reached 78/100 in a century lasted that long, counted from the moment the index enters Elevated until it returns to Normal. The longest ran 59 months.
The same count, split by the calibration cut
This is a range, not an estimate, and it is not a forecast. With 17 episodes in a century, no exact figure would be honest, and more daily data does not create more episodes. It says what past episodes lasted, never how long the current one will.
MSI reads the S&P 500 and nothing else, on purpose: a reading is worth what its calibration is worth, and every market has to earn its own over its own history. That is why the scope is one series, and why widening it is data work rather than new research. Nothing below is built, and none of it carries a date.
One instrument. The S&P 500, calibrated on the first half of the century and checked on the half that followed.
Each one calibrated on its own history, never on a profile borrowed from equities. What every candidate needs is a long daily series, not a different method.
The nearest step, and the one that matters most: the same asset class, read the same way. It is what turns "it works on this series" into "it works on this kind of market".
Decades of public daily closes, and a regime that behaves nothing like equities. Its own calibration, or no reading at all.
Oil and metals move on supply, weather and politics. The honest test of whether a reading built on an index travels at all.
The number already exists. Everything you would put around it is product, and none of it is standing today.
The same 0 to 100 reading across instruments, so tension can be compared instead of read one instrument at a time.
Today a reading is taken. Following it and saying when the state changes is the recurring use, and it is the part that does not exist.
The output is a number and a state. Putting it next to your existing risk measures is integration, not modelling.
Every check repeated on a second provider, so no result rests on a single feed. The tooling for it is written; the run is not done.
No dates, deliberately. A roadmap whose dates slip costs more than having no roadmap at all.
Stated here, by us, rather than in small print at the bottom. A reader who finds an undeclared limit discounts everything else on the page, and is right to.
There is no forecast in the output and no probability attached to what comes next. It reports the state of the market now, and that is the whole claim.
We measured that too, and what it catches of the beginnings is not distinguishable from chance. The reading is about the state of the market, not about the turn.
The reading accumulates instead of resetting each session, which is what lets it show tension that has been building for months. The cost is the fast fall: in 2020 it read 56 during the decline, and 77 in the months after it had ended.
While they are running, an episode that will end in a decline and one the market will correct on its own look alike. What separates them is how long they last, and that is only known afterwards.
The calibration comes from that one series and its own history. Pointed at another market it would treat as unusual what is ordinary there, so it is not pointed at another market.
An episode of tension that the market corrected is not a failed reading. There was tension, and it passed.
Read what we have tested and what it does not show, or start a conversation about where it could be useful.