On 10 Jun 2026 the Nifty option expiring on 30 Jun was priced at an implied volatility of 15.03%. The fifty stocks inside the index, weighted the same way, averaged about 23 percent. Put the two together and the market was pricing the stocks to move together with a correlation of about 0.39. The sessions up to 30 Jun delivered about 0.28. That gap is a correlation risk premium, on one expiry. One expiry proves nothing, so this lesson does two jobs. It shows how to measure the premium and split it from the volatility premium, and it asks what a year of realised correlation says about the level the market priced. Dated NSE data, no live numbers, no view on the next expiry.

3By the end you can

  1. Define implied and realised correlation and the premium between them.
  2. Split an index volatility premium into a stock volatility part and a correlation part.
  3. Say why one expiry cannot establish a premium and which state hurts a short correlation position.
A star field of risk and return dots with one bright frontier arc across it

The idea, plainly

What is being priced

An index option and a stock option are bets on two different things. A stock option prices one stock's moves. The index option prices the weighted sum of all of them, and the variance of a sum depends on how much the parts move together. With a high correlation the stocks pull the index along and its variance is large. With a low one they cancel and the index is calm. So index volatility holds a stock volatility part and a correlation part, and stock options let you read off the first.

Read one more paragraph, 75 words

The correlation that fits the prices on a day is the implied correlation, one average figure for every pair, as built in the implied correlation lesson. The correlation that the returns then show is the realised correlation, found with the same formula but realised volatilities. The correlation risk premium is implied less realised. A positive average means index options were dearer than the stock options beneath them justified, once the realised outcome is known.

Why a premium could exist

Two explanations travel together, and the data here cannot choose between them. The first is demand. Portfolio holders buy index puts to hedge and few sell index options in size, so the index leg is bid up against the stock legs. The second is risk. Correlation rises when markets fall. A seller of index volatility who buys stock volatility is short correlation, and loses in the very state where everything falls together. A premium paid for carrying that state is compensation, not a mispricing. A well known study of US index options reported that correlation risk looked priced and that index options looked dear next to stock options. I have not found a long published test on NSE data, and one expiry cannot stand in for it.

Index hedging demand pushes the index leg up; a sell-off is the price

Where a correlation risk premium could come from: demand for index protection lifts index implied volatility against stock implied volatility, which lifts implied correlation above the correlation that usually follows, and the loss state is a sell-off in which correlation jumpsDiagram with 4 boxes. Hedgers buy index options to Index IV above stock IVs; Index IV above stock IVs to Calm months; Index IV above stock IVs to Sell-off.Hedgers buy index optionsdemandIndex IV above stock IVsimplied correlation risesCalm monthsrealised belowimpliedSell-offcorrelation jumps
The two usual explanations in one chain. Hedgers pay for index protection, the implied level rises, and the party that sells it is short the state in which correlation jumps. A sketch of the argument, not a market number. Source: RupeeCase illustration, after the research named in the sources. As of 17 Oct 2026.

Splitting the index premium in two

The index volatility premium, implied less delivered, is a mix. Part of it is because stock volatilities came in below what stock options priced. The rest is because correlation came in below what the index option priced. Splitting them matters, because the first is a volatility risk premium on each stock and the second is a different risk. The split is done by changing one input at a time, in a fixed order.

The maths: one input at a time

Write the index variance in terms of the weighted average stock vol and the zero correlation end.

s_I^2 = Z^2 + rho x (A^2-Z^2)
A = sum w s
Z^2 = sum (w s)^2

A is the index vol if every stock moved together, and Z the index vol if none did. Step one keeps the implied correlation and swaps in realised stock vols, which gives a middle index vol. Step two swaps in the realised correlation, which lands on the delivered index vol. The first gap is the stock vol part and the second is the correlation part. The order matters a little, so I run it both ways and report the difference.

Worked example: one expiry, then a year

Four readings of the same index

Take the 30 Jun 2026 expiry as it stood on 10 Jun, with the 15.63 India VIX and the index at 23,214.95. The chart gives four correlation readings: what the options implied, what the trailing 60 and 20 sessions had shown, and what the thirteen sessions to expiry delivered.

The implied level sat near the 60 session reading and above what followed

Four correlation readings of the Nifty 50 around 10 Jun 2026: implied 0.39, realised over the 60 sessions before 0.40, over the 20 sessions before 0.18, and over the 13 sessions to the 30 Jun expiry 0.28Implied, 10 Jun 2026 0.39; Realised, 60 sessions before 0.40; Realised, 20 sessions before 0.18; Realised, to 30 Jun 0.28.Average correlation of the Nifty 50, same formula each timeImplied, 10 Jun20260.39Realised, 60sessions before0.40Realised, 20sessions before0.18Realised, to 30Jun0.28
Same formula, weights of 10 Jun 2026, stocks and index from the fifty name file. Realised readings use close to close returns. The weights are a proxy built from fund holdings, not the index provider's own file. Source: NSE end of day data, computed by RupeeCase; weights are a proxy from fund holdings. As of 15 Sep 2026.
Read one more paragraph, 55 words

The implied reading was close to the 60 session one and well above the 20 session one. What came next, over the last stretch before expiry, was in between and lower than the options priced. Notice how much the realised reading depends on the window. That alone says to be careful with a single number.

Where the index premium sat

Now split it. The index option implied 15.03% and the index then delivered 12.8 percent. Replace the stock vols with what they delivered and the index vol barely moves. Replace the correlation as well and it drops to the delivered level.

Almost all of the index premium sat in correlation, not in stock vol

The 2.2 vol point gap between the 15.03 index implied volatility of 10 Jun 2026 and the 12.82 delivered to 30 Jun split into 0.13 from stock volatilities and 2.08 from correlation, so about 94 percent sat in correlationIndex IV, 10 Jun 2026 15.03; Stock vols came in lower −0.13; Correlation came in lower −2.08; Index vol delivered 12.82.Vol points; the middle two bars are changesIndex IV, 10 Jun202615.03Stock vols camein lower-0.13Correlation camein lower-2.08Index voldelivered12.82
Changing one input at a time, stock vols first then correlation. Doing correlation first changes the split by only a couple of hundredths of a vol point, so the picture holds. One expiry, thirteen returns, at the money IVs only. Source: NSE end of day data, computed by RupeeCase. As of 15 Sep 2026.
Read one more paragraph, 81 words

In this one expiry the stock options were priced almost fairly, as the stocks then delivered. Nearly all of the premium was in how the stocks moved together. I would not call that a finding. It is a clean illustration of where to look, on a sample far too small to bank on. Thirteen daily returns give a noisy volatility for the index and for each stock, and at the money IVs leave out skew, which is steeper on the index.

One reading against a year of realised ones

A single implied reading needs a distribution to be compared with. The file holds 241 sessions of 20 session realised correlation, from 24 Sep 2025 to 15 Sep 2026, on the same weights and the same formula. It ranged from 0.035 to 0.602, with a mean of 0.252 and a median of 0.234.

Realised correlation sat below the implied level on about seven sessions in eight

Count of 241 sessions from 24 Sep 2025 to 15 Sep 2026 by 20 session realised correlation of the Nifty 50: 11 below 0.10, 76 from 0.10 to 0.20, 102 from 0.20 to 0.30, 23 from 0.30 to 0.40, 10 from 0.40 to 0.50 and 19 at 0.50 and above, with the implied 0.39 of 10 Jun 2026 in the 0.30 to 0.40 groupBelow 0.10 11; 0.10 to 0.20 76; 0.20 to 0.30 102; 0.30 to 0.40 23; 0.40 to 0.50 10; 0.50 and above 19.Sessions by 20 session realised correlation, 241 in allBelow 0.10110.10 to 0.20760.20 to 0.301020.30 to 0.40230.40 to 0.50100.50 and above19
Sessions grouped by 20 session realised correlation, weights of 10 Jun 2026 held fixed on every date, so the series is a proxy that drifts as true weights change. The marked group holds the implied correlation of one expiry, not a series. Source: NSE end of day data, computed by RupeeCase; weights are a proxy from fund holdings. As of 15 Sep 2026.
Read one more paragraph, 68 words

The implied level was above the realised reading on about seven sessions in eight. That is the fact a premium trader would like, and it is a weaker one than it looks. The implied level here is one date. I have no history of implied correlation, so I cannot say whether that level was high or low for implied. I can only say it was high for realised.

Correlation climbs with fear

The risk side shows up in the same file. Sort the sessions by India VIX into four equal groups and average the 20 session realised correlation in each.

In the highest VIX quarter, stocks moved together about twice as much

Mean 20 session realised correlation of the Nifty 50 by India VIX quarter over 241 sessions to 15 Sep 2026: 0.19, 0.19 and 0.23 in the three calmer quarters and 0.40 in the highestVIX up to 11.4 0.19; VIX 11.4 to 12.6 0.19; VIX 12.6 to 15.9 0.23; VIX above 15.9 0.40.Mean 20 session realised correlation, by quarter of sessionsVIX up to 11.40.19VIX 11.4 to 12.60.19VIX 12.6 to 15.90.23VIX above 15.90.40
Sessions from 24 Sep 2025 to 15 Sep 2026 sorted by India VIX into four groups of about the same size. India VIX is a 30 day expectation, the correlation a trailing one, so they share months, not days. One year, one stress episode. Source: NSE end of day data, computed by RupeeCase. As of 15 Sep 2026.
Read one more paragraph, 77 words

In the three calmer quarters the stocks moved together about half as much as in the highest one. The implied level of 10 Jun sat at roughly what the stressed quarter delivered. Read that as the market charging for the stress state, not as a forecast of it. It also shows how a short correlation position loses. It does not lose a little each month. It loses when the state changes, and the state changes with volatility.

How I would measure it properly

One expiry is a single draw. A real estimate needs many, and each choice below changes the answer.

  • Same maturity. The index option and the stock options must expire on the same day, so the monthly series is the natural choice. Weekly Nifty options have no weekly stock partners.
  • Same strike basis. Use the at the money IV or a fixed delta point for every name and the index. Mix them and skew leaks into the correlation.
  • Weights from the index provider. My weights are a proxy from fund holdings. They move the answer a little each month.
  • A realised window that matches the option. The trailing window is what was known on the day. The window to expiry is what the option was priced against. Report the second for the premium.
  • Many expiries, none overlapping. Monthly expiries give twelve a year. Even three years is about three dozen draws, and the standard error will still be wide.

I also keep the stock vol part and the correlation part separate in every table, so a reader can see which one a premium came from. A single combined number hides the question this lesson is about.

Appendix: how noisy is a volatility from a few returns?

For a volatility estimated from n daily returns, the relative standard error under normal returns is roughly the figure below. It is the reason a thirteen return window gives a fuzzy realised correlation.

SE(s) / s = 1 / sqrt(2(n-1))

With n returns of a dozen or so, the relative error is about a fifth. A 20 session window narrows it to about a sixth, and a 60 session window to about a tenth. Correlation is built from these volatilities, so its error is larger still, and stock returns are not normal.

The term this section turns on: Realised correlation

The term this section turns on: Realised correlationThe term this section turns on: Realised correlation. Realised correlation: The same formula with realised volatilities of the stocks and the index over a window of returns.Realised correlationThe same formula with realised volatilities ofthe stocks and the index over a window ofreturns.
From the module glossary; every term is in the Learn glossary too. Source: Module 18.12, The correlation risk premium.

Mapping correlation to index volatility

The last piece is arithmetic. Holding the stocks at the volatilities they delivered, each assumed correlation gives an index volatility. The index option broke even, delivered against implied, at a realised correlation very close to the implied one, as it must when the stock vols come in about right.

Appendix: the index vol at each correlation

The index option breaks even where the delivered stock vols meet its price

Index volatility implied by each assumed correlation when the 50 stocks have the volatilities they delivered from 10 to 30 Jun 2026: 4.70 with none, 12.82 at the delivered 0.28, 15.03 at the break-even 0.40 and 23.11 with all togetherNo correlation 4.70; Correlation 0.10 8.56; Correlation 0.20 11.16; 0.28, delivered 12.82; 0.40, break-even 15.03; Correlation 0.60 18.15; All together 23.11.Index vol, percent, at each assumed correlationNo correlation4.70Correlation 0.108.56Correlation 0.2011.160.28, delivered12.820.40, break-even15.03Correlation 0.6018.15All together23.11
The index volatility that each assumed average correlation gives, holding the stocks at the volatilities they delivered from 10 to 30 Jun 2026 and the weights of 10 Jun. The marked bars are the delivered correlation and the break-even one. Source: NSE end of day data, computed by RupeeCase. As of 15 Sep 2026.

In variance terms the relation is a straight line, which is why a correlation view is easy to write down. In vol points the curve flattens, so the same gain in correlation adds less volatility the higher it starts. Variance is what a variance swap pays, and the variance risk premium lesson shows how to read it.

What India adds

There is no listed correlation or variance contract in India, so the premium cannot be bought or sold directly. The only way to hold the view is the dispersion trade, which the dispersion lesson builds: index options against many stock options. Stock options expire monthly and settle by delivery, while index options settle in cash, and stock option spreads are wider than index ones. A paper premium of two vol points is exposed to all of that before it is exposed to correlation.

Read one more paragraph, 96 words

Capital and costs add more. Fifty stock legs mean fifty margin blocks, fifty sets of statutory charges and fifty spreads to cross, against one on the index. A stock can enter the ban list or have its lot size revised from a new series, and a stock option that ends in the money turns into a delivery of shares. The premium has to clear all of that every month, in the months that go wrong as well as the ones that go right. I treat a premium I cannot cost as a premium I cannot count.

The common mistake

The claim: Implied correlation is almost always above realised, so a dispersion trade is a free premium. A positive average is evidence that sellers of index volatility were paid for a risk, and the risk is a state in which correlation jumps. In this file the stocks moved together about twice as much in the highest VIX quarter as in the calmer ones, and the implied level of one expiry was near that stressed reading. One expiry shows where to look, not what it pays. Before accepting a premium, ask for the history of implied, the window of realised, the weights, and the state in which it hurts.

The RupeeCase terminal on sample data: Model history or your portfolio: which numbers am I looking at?
Watch in the terminal: which numbers am I looking at? A short tutorial on sample data.

Key takeaways

  • The premium is implied correlation less the correlation that followed.
  • Index volatility splits into a stock volatility part and a correlation part.
  • In one expiry almost all of the index premium sat in correlation.
  • Stocks moved together about twice as much in the highest VIX quarter.
  • One expiry shows where to look, not what the premium pays.
  • India has no listed correlation contract, so the view needs many option legs.

Two terms to hold on to from 18.12

Two terms to hold on to from 18.12Two terms to hold on to from 18.12. Implied correlation: The single average correlation between every pair of stocks that makes the weighted stock IVs add up to the index IV on a date. Correlation risk premium: Implied correlation less the correlation that followed, positive when index options were dearer than their stock options justified.Implied correlationThe single average correlation between everypair of stocks that makes the weighted stockIVs add up to the index IV on a date.Correlation risk premiumImplied correlation less the correlation thatfollowed, positive when index options weredearer than their stock options justified.
From the module glossary. Source: Module 18.12, The correlation risk premium.

Before you trade. Derivatives trading carries a high risk of loss. SEBI's study "Profitability of Individual Traders in the Equity Derivatives Segment (FY25 to FY26)" (August 2026) found that 87.7% of individual traders in equity F&O made net losses in FY26, with aggregate net losses of Rs 91,685 crore. Read the SEBI study