Sector Concentration: The Correlation Risk in a Put-Selling Book

Counting tickers is not measuring risk

A put seller with eight open contracts across eight different companies will usually describe that as a diversified book. The reasoning is intuitive: no single company failure can do much damage, because no single company is more than an eighth of the exposure.

The reasoning is sound for the risk it addresses and silent about the one that matters. Idiosyncratic risk — the accounting scandal, the failed trial, the lost contract — is genuinely reduced by spreading across names. That is the risk that arrives on a random Tuesday and affects one company.

The risk that takes a put-selling book apart is the other kind: the one that arrives on a specific morning and affects everything at once. Against that, eight tickers can behave as one position, and the count offers no protection at all.

Why a ticker count is the wrong measure

Take a book of eight short puts on eight companies, each sized identically. Ask what fraction of it is at risk from a single event.

Against a company-specific event — a failed trial, a lost contract, an accounting problem — the answer is one eighth. That is what diversification bought you, and it is real.

Against a market-wide fall, the answer is close to all of it. Every underlying is an equity. On a day when equities drop three percent, all eight strikes move toward the money together, and the ones that were closest cross it together. Nothing about holding eight tickers instead of one changes that, because the driver is not company-specific.

The uncomfortable part is that both statements are true of the same book at the same time. It is genuinely diversified against the risk that shows up in a normal month, and barely diversified at all against the risk that does the damage. A ticker count measures the first and is silent on the second.

The clustering this produces

Because the driver is shared, the outcomes are not evenly spaced. A put-selling book tends to run for long stretches with losses arriving one at a time, and then resolve a large share of its positions against itself on a single morning.

That distribution matters more than the average rate, and the average conceals it. A loss rate of about one in twelve does not mean one contract in every twelve fails in turn; it means most expirations pass with nothing, and occasional ones take many at once. A seller who has planned for the average has planned for a pattern their book will not follow.

The same effect appears in every published record of a short-premium strategy, including across the 2018, 2020 and 2025 episodes covered in short puts in a market-wide drawdown. It is a property of selling correlated insurance, not a feature of any one engine or any one year.

Where the hidden correlation comes from

Sector. The obvious one. Four regional banks are one bet on rates and credit. Three semiconductor names are one bet on the cycle. The tickers differ and the driver does not.

Factor. Less obvious and more common. High-multiple growth names across unrelated sectors move together on a rate surprise. Small caps move together on liquidity. Anything with a heavy overseas revenue share moves together on the dollar. A book can look diversified by sector and be a single factor bet.

Volatility itself. The most easily missed. A screen that selects on elevated implied volatility will systematically return names that are volatile at the same time, because whatever raised volatility across the market raised it for all of them. Selecting on the characteristic that makes premium rich also selects for names that respond to the same conditions. The screening criterion creates the correlation.

Expiration date. Structural and almost never counted. Positions sharing an expiration share a single settlement, and everything is decided by where prices happen to sit on one morning. This is the clearest case of the whole problem: not many independent outcomes, but one market day resolving all of them simultaneously.

The interaction that does the damage

Correlated assignment is worse than correlated marks, because the consequences compound.

Several contracts assigned on the same morning consume several allocations of cash at once, and the cash was set aside on the assumption that assignments would be occasional and staggered. A book that could comfortably absorb one assignment a month may not be able to absorb five on one Monday, and it discovers this after the fact.

The positions that arrive are also correlated with each other, so the seller is not left holding a diversified equity portfolio. They are left holding a concentrated one, acquired at above-market prices, in whatever was falling. And having spent the cash, they cannot write new contracts, which removes the income the strategy exists to produce at precisely the point they would most like to be earning it.

None of these consequences is visible when the positions are opened one at a time, each looking reasonable on its own.

Measuring it without a risk system

Three checks, none requiring a covariance matrix.

The single-factor question. Name the thing that would hurt the most positions at once — a rate move, a dollar move, a growth-to-value rotation, a broad risk-off day. Then count how many positions it touches. If the honest answer is most of them, the book is one position however many tickers are in it.

The shared-expiry count. Group open contracts by expiration date and look at the largest group. That is the number of positions that will be decided by a single morning's prices. If it is most of the book, the settlement risk is concentrated regardless of how the underlyings are spread.

The simultaneous-assignment test. Take the largest correlated group, assume every contract in it is assigned on the same day, and total the cash required. If that number exceeds what is available, or leaves nothing for anything else, the book is already too concentrated. This is the version of the question that matters, because it is the one a bad morning asks.

What actually reduces it

Stagger expirations deliberately. Spreading across expiration dates is the cheapest diversification available to a put seller and the most consistently neglected. It converts one settlement into several, which is the difference between one bad morning and several ordinary ones.

Cap correlated exposure, not ticker count. A limit expressed as a share of the book responding to one factor is a real constraint. A limit on positions per ticker is not, because it permits the thing that goes wrong.

Treat volatility clustering as information about the book. When a screen starts returning many names at once with unusually rich premium, that is information about market conditions rather than an unusual crop of individually good contracts. The right response to abundant premium is usually smaller size, not more positions.

How StratosIQ handles it, and what it does not do

Being precise about the boundary matters here.

The model scores contracts. It applies its safety floor to each one independently, and a contract failing that floor does not publish however rich its premium looks. What it does not do is manage a portfolio, know what you already hold, or apply a concentration limit across your book. It has no view of your other positions, and it cannot have one.

That boundary is deliberate rather than a missing feature. Concentration is a property of a portfolio, and the portfolio is yours. A screening tool that implied otherwise would be making a claim it has no basis for.

One thing does help at the source. Patrol excludes contracts whose expiration spans an earnings announcement, which removes a category of clustered event risk from the steady-state book. Strike takes the opposite condition deliberately, working relative to the announcement date. Knowing which engine a contract came from tells you which kind of event exposure it carries.

The full published record, contract by contract and including every loss, is on the performance page.

Worth reading next

Short puts in a market-wide drawdown is the extended version of the same problem, running over weeks rather than one settlement. Why a 90% win rate can still lose money covers why clustered losses hit expectancy harder than the per-contract arithmetic suggests.