On 10 Jun 2026 the Nifty's at the money implied volatility for the 30 Jun expiry was 15.03%. Two of the three heaviest stocks in the index were at 23.52% and 26.69% for the same expiry. A weighted average over all fifty names was also well above the index figure. The index was priced far calmer than its parts. That gap is what correlation means in option prices. If all fifty stocks moved together, the index would be as volatile as the average stock. If they moved independently, it would be far calmer. The market quoted something between. This lesson turns the gap into one number, sets it against what the stocks then did, and lists what the number leaves out.
3By the end you can
- Compute the average implied correlation of an index from its volatility, its stock volatilities and weights.
- Place the index between its no correlation floor and full correlation ceiling, and test the answer against realised correlation.
- Say which inputs move the number most and which checks catch a bad one.
The idea, plainly
Index variance, in parts
The variance of a portfolio is the sum of two kinds of terms. One is each stock's own variance, scaled by its weight squared. The other is every ordered pair of different stocks, i and j with i not equal to j, so each pair counts twice, scaled by both weights, both volatilities and the correlation of the pair.
Index variance = Σ w2s2
+ Σi≠j wiwjsisj x rhoij
Read one more paragraph, 44 words
The index has one implied volatility, from its own options. Each stock has one, from its own. The weights are known. The pair correlations are not quoted anywhere. There are 1,225 pairs among fifty names, and no option trades on any one of them.
One average correlation
So replace every pair's correlation with one number, the same for all pairs, and solve for it. That number is the average implied correlation.
rho = (I2 − Σ w2s2)
/ ((Σ w s)2 − Σ w2s2)
The maths: where the formula comes from
Set every pair correlation to the same rho. Then the pair terms are rho times the sum over ordered pairs, i not equal to j, of wiwjsisj. That sum equals the square of the sum of the weighted volatilities minus the diagonal.
Sum over i≠j =
(Σ w s)2 − Σ w2s2
Write A for (Σ w s) squared and B for the sum of squared weighted volatilities. Then index variance is B plus rho times (A minus B). Solving for rho gives the formula above. At rho equal to zero the index volatility is the square root of B. At rho equal to one it is the square root of A, which is the plain weighted average of the stock volatilities. The formula does not force rho into the range from minus one to one: bad inputs can push it outside, and a result outside is a sign of bad inputs, not of a finding.
For small moves, the derivative of rho with respect to the index volatility I is two I over (A minus B). With A minus B of about five hundredths and I near fifteen hundredths, one volatility point on the index moves rho by about six hundredths.
Read one more paragraph, 59 words
Here I is the index volatility, s a stock's volatility and w its weight, all as decimals and all for the same expiry. The numerator is how much of the index variance the stocks' own terms do not explain. The denominator is the most that the pair terms could add, which is what happens if every correlation is one.
The two ends of the range
Because of that algebra, the index volatility has a floor and a ceiling for given stock volatilities. The floor is the square root of B, the index if no pair moved together. The ceiling is the weighted average of the stock volatilities, the index if every pair moved perfectly together. The implied correlation says where the index sits between them, in variance terms.
2 terms this section turns on
What goes in
- The index volatility. The at the money volatility of the monthly expiry, a shortcut for the full smile.
- The stock volatilities. The at the money volatility of each stock's option for the same expiry. Stock options expire monthly and settle by physical delivery, so their prices carry features that index options lack.
- The weights. The index weights, which come from free float market value by NSE Indices' method. This lesson uses a proxy: an index fund's month-end portfolio statement of 31 May 2026, valued at the prices of the day. It is not NSE Indices' own weight file, which is the better input.
- The date. End of day settlement prices, the same day for every option.
Four ways it can move
The formula has two sides, so the number can move for four reasons that look different on a screen.
- Index volatility rises, stock volatilities do not. Implied correlation rises. Buyers are paying for more joint movement than the single names show.
- Stock volatilities rise, the index does not. Implied correlation falls. The risk is in single names, and it is averaging out inside the index.
- Both rise together. The two effects partly offset. On 10 Jun 2026 a point on both raised the number by about two hundredths, a third of the index effect alone. A broad rise in fear can leave it close to where it was.
- The weights change. A heavier top name pushes the floor up. An index fund's statement is a month old, and an index rebalance can shift the weights between statements.
So a rise in the number is not the same as a rise in fear, and the other way round. Read the three inputs, not just the output.
Why anyone computes it
Because it is a price. Index options are often bought as insurance against a fall in the whole market. Stock options are bought and sold for many reasons, mostly about one company. A published study of US index options, listed in the sources, found that correlation risk is priced, with sellers of index volatility earning a premium on average. That is a claim about one market and one period. Module E09 takes it up. Whether the same holds for the Nifty is a question for a long sample, and the one expiry here cannot answer it.
Checking the inputs
Most surprises in this calculation are input errors. Check that every option is for the same expiry and the same date, that each stock volatility came from a settlement price with traded volume, and that the weights add to one after rescaling. Check that no stock is in a window where it had a corporate action: a split looks like a crash in a close to close series. In this data set five names were flagged for that reason, and they are left out of the windows that hold the jump. Then check that the answer is between the floor and the ceiling.
Read one more paragraph, 61 words
There is no Bank Nifty version in this lesson. The data set has no weights file for its stocks, and the banks that are also in the Nifty 50 are not a basket of the whole index. The formula is the same for that index, and a concentrated index with fewer names gives a quicker calculation and a more sensitive answer.
What it is not
It is not any one pair's correlation. It is one average standing in for all pairs. It is not the realised correlation either: it is the number that the option prices imply, which includes whatever risk premium sellers charge. It does not say that realised correlation will fall or rise. And the at the money shortcut ignores skew. The index smile and a single stock's smile have different shapes, so reading one strike off each is a convention that this lesson uses, not a law of nature. A method that uses a strip of strikes is a refinement.
Worked example: fifty names, 10 Jun 2026, 30 Jun expiry
The index volatility from the Nifty chain is 15.03%. The fifty weights add to a hundred. The heaviest is 10.78% of the index. Put each stock's at the money volatility beside its weight and compute B and A. Then the two ends and the index sit as follows.
The index is priced between its two ends, and nearer the floor
Implied correlation sat above the recent past and level with the last quarter
Realised correlation usually sat below the implied level, and spiked once
Read 2 more paragraphs, 126 words
Solve the formula and the implied correlation comes out at the value shown in the next chart, against three realised readings taken the same way. The realised readings use the same formula with each name's realised volatility and the index's realised volatility, over the window shown, holding the same weights.
The twenty session reading before the date was 0.18. The sixty session reading is in the chart above. The implied figure was above the first and level with the second. Over the thirteen sessions to the expiry the realised figure was lower than the implied one. That is one expiry, one draw. Across the series, from 24 Sep 2025 to 15 Sep 2026, the twenty session reading averaged 0.25 and ranged up to 0.60.
How many names do you need
Fifty is a lot of options. How far does the answer move if you keep only the heaviest names and renormalise? Compute the implied correlation again with the top five, ten, twenty, thirty and all fifty names, using the same index volatility.
Ten names give almost the answer that fifty give
Read one more paragraph, 57 words
With five names the figure drops. From ten names on it settles. That matters in practice, since a basket of ten options is far easier to run than a basket of fifty. It also matters for honesty: the figure is not very sensitive to the tail of small weights, and it is very sensitive to something else.
The input it cares about most
Move the index volatility by one point, or every stock volatility by one point, and see what the correlation does.
One volatility point on the index moves the correlation by about six hundredths
Appendix: the same arithmetic, step by step
List the fifty names with their weights and at the money volatilities. Rescale the weights to add to one. Compute w times s for each name and add them: that is the weighted average volatility, the ceiling. Compute the sum of squares of w times s, B, and take its square root: that is the floor. Square the index volatility and subtract B. Divide by the square of the ceiling minus B. Check that the result lies between minus one and one, and that the index volatility lies between the floor and the ceiling. If it does not, look for a stale price, a wrong expiry or a weight that has not been normalised before looking for a market story. The tool beside this lesson does all of it and lets you change the inputs.
Read one more paragraph, 64 words
A point of volatility is a small move in a quote, and a few hundredths of correlation is a big move in the number we are reading. Option prices have a bid and an ask, and a settlement price has its own rounding. So the third decimal of an implied correlation says nothing, and the second is often noise. Read it to one decimal.
The common mistake
The claim: Implied correlation is above realised correlation, so correlation is overpriced and will fall.
On 10 Jun 2026 the implied figure was above the twenty session reading and level with the sixty session one, and the realised figure over the next thirteen sessions, a short window, was lower. One expiry fits the claim. But the same series reached 0.60 on 8 Apr 2026, when index volatility spiked. A single expiry is one draw from a spread of outcomes. Implied above realised describes a price paid for protection, which is not a forecast. Ask how often it held, over how many expiries, and what the bad draws cost.
Key takeaways
- Index variance splits into stock variances and the correlations between stocks.
- Implied correlation solves that split for one average correlation.
- It places the index between a no correlation floor and a full correlation ceiling.
- Implied correlation is sensitive: one index volatility point moves it noticeably.
- Ten heavy names give almost the same answer as all fifty.
- Implied beside realised correlation on one expiry is a comparison, not a signal.
Two terms to hold on to from 18.10
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