The Phantom Close

Charts started doing something impossible in the last two minutes of the day.

On 3 August 2026, India moved to a closing auction for stocks with derivatives. Instead of the close being the last traded price, orders now pool in a 20-minute auction and match at a single equilibrium price. It is a well-established mechanism — most major markets use one.

What was not anticipated is what it did to intraday charts.

Because a headline index is calculated from its constituents' closing prices, the index cannot update until those auction closes are published. So the index goes completely flat for roughly fourteen minutes — then moves its entire day's remaining distance in a single one-minute bar.

On some days that final bar was larger than the entire preceding session. Anyone running a chart type that reacts to price movement rather than to clock time — Renko, Point & Figure, range bars — saw a burst of activity that never happened. On the worst day, roughly a third of the entire day's chart events printed inside those two minutes.

Compare what the index printed against what was actually trading.

The index is a computed number. But two things kept trading continuously right through the auction window, and both are real markets with real money behind them:

• Index futures — never stop, never freeze, real volume throughout.
• Index options — also continuous, and they carry an independent read on fair value.

The options are the stronger test, because they can be interrogated without any pricing model at all. Put-call parity says that for the same strike and expiry, a call price minus a put price implies the forward level directly: C − P = F − K. No volatility assumption, no delta, no model. Just arithmetic on two quoted prices.

So for each session: freeze the clock at the last continuous minute, then read the print minute three ways — what the index said, what the future did, and what the option market implied.

The print is real as a closing value. It is fiction as a price.

Across every non-expiry session measured, the pattern held without exception. The index printed tens to hundreds of points. The instruments that were actually trading moved a handful.

Two different strikes, tested independently on the same session, implied +14.25 and +14.9 — so this is a consistent read, not an artefact of one contract.

The one exception proves the mechanism. On expiry day the option market did converge toward the print — because index options settle to precisely that auction-derived close. Where the print is economically binding, the market prices it. Everywhere else, it ignores it.

Which gives the clean statement of the problem:

And the freeze is worse than the jump.

It is tempting to describe this as a phantom bar appearing out of nowhere. The options market says otherwise.

Take an expiry session, where the auction close decides whether a contract pays or expires worthless. A put 37 points in-the-money at the last continuous minute, quoted at 56.35, settled worthless — the auction closed above its strike. The mirror-image call went from 48.05 to a 114.70 settlement.

Now look at the middle column. The put was already down to 14.65 before the print landed. The option market had been repricing continuously through the freeze — the exchange publishes an indicative auction price, and traders were trading on it. Priced off those options, the implied move was already most of the way to the eventual answer.

Which means the outcome was largely visible in advance — to everyone except someone watching the index.

This is worth being precise about, because it is easy to reach for the wrong fix. The print is not bad data. It is not an error to be filtered. It is a correct closing value for the purpose it was designed for — settling contracts. The mismatch is that a settlement value is being fed into tools that assume every bar represents trading.

And there is a second-order effect that matters more than the first. A phantom bar does not just distort the day it lands on. Movement-based charts carry state forward: bricks and boxes persist. A burst of manufactured bars shifts the chart's structure, and every trend reading computed afterwards inherits that shift. The distortion outlives the session that produced it — which is why "it was only two minutes" understates it considerably.

Four routes, and the useful one is not the obvious one.

Anyone hitting this will consider roughly the same set. They are not equally good.

Before you reach for futures — check that yours actually trades.

Both futures-based routes assume a contract that prints continuously. That assumption is safe on a headline large-cap index and can be badly wrong one rung down the liquidity ladder. Measured on a broad-market mid-cap index against its large-cap sibling — share of session minutes in which the near-month future recorded no trade at all:

The fourth route works because the basis between index and futures is stable and slow-moving over a window this short — a drift measured in fractions of a point across twenty-five minutes. Carrying it forward across the freeze is well inside the tolerance of any sensible chart.

One honest caveat, worth stating rather than burying. Replacing a 200-point phantom with an 11-point estimate plainly removes a fiction. Whether that estimate is the best available reconstruction of fair value is a separate question, and a harder one — the sample of days where it can be tested cleanly is still small. The right sequence is to measure that before building on it.

The thing we found by accident cost more than the thing we went looking for.

This deserves saying plainly, because it is the kind of error that hides behind a market-structure story: the change described above was not the expensive problem. While measuring it, a much larger one surfaced, and it has nothing to do with the auction.

Take a common defined-risk position: sell an at-the-money option, buy one a fixed distance further out — say 300 points — as the hedge that caps the loss. Simple, mechanical, easy to automate. The width is chosen once and left alone.

Over four months, realized volatility on one mid-cap index fell from 22% to 7.7%. Nothing about the position changed. But the meaning of "300 points" did:

Two practical consequences. If strikes are chosen to collect a target premium, falling volatility drags them toward the money automatically — the structure gets riskier without a decision being taken. And if the width is fixed in points, the protection thins out on exactly the same schedule. Either way the drift is silent, and it shows up as "the same move costs more than it used to."

Worth checking on your own book before blaming anything structural: express your hedge distance in standard deviations rather than points, and see where it has drifted to.