OI Cohort Exits: Who Is Unwinding?

Most option-chain tools tell you that open interest fell. Far fewer try to tell you who unwound it and why. Nakshatra’s OI cohort-exit model is its signature feature: a direction-aware inference that takes an OI drop apart and estimates whether it was sellers buying back in pain or buyers liquidating, broken down by when those positions were likely opened. It is genuinely useful — and it is an estimate, which this page is careful to be honest about.

The problem with a bare OI drop

When the open interest at a strike falls, something closed. But a bare OI decline is ambiguous. Was it the writers (short side) covering because the trade went against them? Was it the buyers (long side) liquidating? And were those positions opened today, or are they leftovers from days ago carrying very different cost bases? A single number — “OI down 40,000” — answers none of this, yet the answers change what the move means.

The exchange does not publish position holders, so no tool can read this off a feed. What it can do is reason carefully from what is observable: the sequence of OI changes, and how the premium moved relative to what the spot move alone would explain. That reasoning is what the cohort model formalises.

Cohorts: grouping OI by when it was built

The model first groups the buildup into cohorts by IST calendar date — roughly, “positions that appear to have been opened on day X.” Because OI accumulates over time and Nakshatra stores every 5-minute snapshot, it can attribute fresh OI to the day it appeared and track that cohort forward.

There is one position the model cannot date: open interest that already existed before our data series begins. That OI has no observable entry day, so it is placed in a flagged carry-over cohort. This is a deliberate honesty mechanism — rather than guessing an origin for positions it never saw open, the model labels them as carry-over so you know their cost basis is unknown.

Delta-adjusting the premium move

The heart of the model is classifying each OI drop by how the premium behaved as the position closed — but the raw premium move is misleading. On a trending day, an option’s price moves largely because spot moved, not because sentiment changed. Reading raw last-traded-price direction therefore misfires.

So the model delta-adjusts. Using the strike’s own implied volatility from the chain, it computes a Black-Scholes delta and removes the spot-explained portion of the premium move. What remains is the part of the move that reflects actual repricing — the signal that hints at who was pressing the trade. Each OI drop is then classified into one of:

• short_covering_pressure — consistent with writers buying back under pain.
• long_liquidation_or_mixed — consistent with buyers exiting, possibly muddied.
• mixed_unwind — evidence does not point cleanly to one side.
• overnight_adjustment — gap-driven repositioning across a session boundary.

The “mixed” categories are not a cop-out; they are used honestly whenever the delta-adjusted evidence is genuinely ambiguous, rather than forcing a confident label the data does not support.

For each cohort, the model matches the classification of its OI drops to seller-pain or buyer-pain and reports a share between 0 and 1 describing how that cohort’s exit breaks down. The output is presented as shares, not raw unit counts, so you read it as “this cohort’s unwind looks roughly 70% seller-pain” rather than as a spuriously precise contract tally.

And it comes with a confidence level, because every step — the cohort dating, the delta adjustment, the classification — is an inference layered on observable data. High confidence means the OI moves and delta-adjusted repricing line up cleanly; low confidence means the picture is murky and you should lean on the label only lightly. The honest framing is essential: this is the most defensible estimate the model can make of who is exiting, not a certified ledger of position holders.

How Nakshatra shows this

The OI Cohort Exits panel lives on the Insights tab. Pick a strike and side and it lays out the cohorts — each entry-day group plus the flagged carry-over cohort — with each OI drop’s classification, the seller-pain versus buyer-pain share, and the confidence behind the read. The Option Chain page shows a clickable notice pointing here, since the analysis belongs alongside the other Insights signals. Read it for what it is: a transparent, direction-aware estimate of who is unwinding a strike — the kind of question most tools do not even attempt — with its uncertainty stated up front rather than hidden.

See this live in the Nakshatra tool →