Guide · Options and volatility · Updated 2026-09-23

India VIX usually sits above the volatility that follows, because someone is paid to carry the month when it does not

The usual chart puts India VIX beside a line of historical volatility from the last month and reads the space between them as the premium. That compares a price for the next thirty days with a measurement of the previous thirty. On this record the backward comparison shows the index above realised movement on 86.5 per cent of days rather than 82.6, and its worst reading is -37.57 rather than -63.71. It hides 41 per cent of the worst gap, because a backward window absorbs a crash only after the index has already repriced it.

Everything below is computed from the exchange's own session files, with the estimator stated so the arithmetic can be redone. Each day's index is paired with the thirty calendar days after it, on the clock the index itself uses, because that is the only pairing in which the gap has a meaning.

A price for variance, built in one specific way

Implied volatility is not measured from anything. It is the one input to an option price that cannot be observed, so it is solved for from the price, as the implied volatility guide works through from the model up. That makes it a price, and a price contains whatever buyers will pay above the outcome they expect.

India VIX is one construction of that price, and the construction decides what it can fairly be compared with. The exchange's methodology note and white paper specify it: the best bid and ask quotes of out of the money options in the near and next monthly books; any strike whose spread is wider than 30 per cent of its mid, or that has no quote, replaced by a natural cubic spline through the good strikes; each option weighted by its strike interval over the square of its strike; the forward taken from the futures of the same expiry and the at the money strike set at the first strike below it; and the two resulting variances interpolated to a constant thirty calendar days on a clock that counts minutes, 43,200 of them in the window and 525,600 in the year. With three trading days left in the near book the index rolls to the next and far books.

The weighting is what matters here. Weighting each out of the money price by the inverse square of its strike makes the sum a variance, the expected squared move across the whole distribution, not an average of volatilities at a few strikes. India VIX squared is therefore a price for variance over the next thirty calendar days, close to the rate at which a variance swap on the index would be struck. The like for like realised quantity is the sum of squared daily returns over those same thirty calendar days, scaled by 365 over 30, with no mean subtracted. The India VIX guide walks the formula strike by strike.

The construction can be checked rather than trusted, because the exchange's end of day files carry closing prices for both monthly books and their futures. Running the published formula on them, a strike counts as a good quote when it traded that session, strikes between traded ones are filled by the same natural spline, the rate is a fixed 6 per cent, and sessions inside the roll window are left out because the panel holds no far book to roll into.

Closing prices are not mid quotes and the rate is a stand in, and the rebuild still lands within a median +0.06 points of the published close. The last two rows are the useful ones. Interpolated the same way, at the money implied volatility sat below the published index on 99.3 per cent of sessions, by a median 0.98 points. That difference is the wings: the strip includes out of the money puts priced for a crash, which a seller of at the money options never sells. A premium measured against the published index is about a point larger than the one an at the money seller is offered.

Realised volatility is an estimator, and every choice in it moves the answer

Realised volatility sounds like a fact and is an estimate with at least four decisions inside it. The return: close to close in logs, which includes the overnight gap. That matters, because over the period India VIX covers, the squared moves from one close to the next open summed to 39.4 per cent of the squared close to close moves, all of it arriving while no position can be adjusted; an estimator built only from the session's high and low misses that part, and an option seller carries it. The centre: a sample standard deviation subtracts the window's own mean, which a variance price does not. The window: thirty calendar days held between 15 and 23 sessions across the record, so a fixed 21 sessions is a different window. The clock: 252 sessions a year, or the index's own 365 day year applied to the window's total.

The four forward estimators agree closely: means from +2.46 to +2.70, and the index above realised on 79.9 to 82.6 per cent of days. Centring and clock move the headline by a few tenths of a point and a couple of points of share, enough for two honest pages to disagree without either being wrong, which is why a figure without its estimator cannot be checked. The backward estimators answer a different question. They put the index above realised movement more often and with a milder worst case, because they set a price against a month that has already happened. The same thirty days can sit on either side: the window that opened on 19 August 2026 realised 7.91, and on 18 September 2026 the backward estimator reads 7.91, because it is the same month seen from its other end.

Thousands of small positive gaps and a handful of enormous negative ones

Across 3,027 windows, India VIX exceeded the volatility that followed on 2,500 of them, 82.6 per cent. The mean gap was +2.70 points, the median +3.30, the standard deviation 6.26 and the skewness -4.69. None of that is a win rate. No position is being scored, only a price against the quantity it priced.

The shape is the finding. The positive gaps are small and numerous, averaging +4.44; the negative ones are fewer and larger, averaging -5.57, and the most negative, -63.71, is twice the size of the most positive, +31.11. Measured in variance, the unit the index is built in, the asymmetry doubles: the windows in which realised exceeded implied sum to 50 per cent of the windows in which implied exceeded realised, against 26 per cent in volatility points. The worst 30 windows, one per cent of the record and 23 of them opened in 2020, carry 60.7 per cent of all the negative variance.

Read the right hand column. In the median window realised variance was 0.59 of the variance priced, so 41 per cent of what was paid for never arrived. Summed across every window, realised variance came to 81 per cent of the variance priced, so the share that never arrived falls from 41 per cent in the median window to 19 per cent in aggregate: the heavy windows take back 54 per cent of what the typical one leaves. One window in ten realised more than it priced; one in a hundred realised 3.4 times as much; the worst realised 22.8 times.

Overlapping windows share most of their days, so the count of windows overstates the independent evidence. Keeping only the first window opened in each calendar month leaves 148 windows that do not overlap, and the shape survives: the index was above realised on 83.8 per cent of them, the median gap was +3.37, and the worst opened on 2 March 2020 at -58.01.

February 2020: one window realised 22.8 times the variance priced

On 17 January 2020 the index read 14.13 and the thirty days that followed realised 15.42, an ordinary month. On 20 February 2020 it read 13.70, slightly lower, and the thirty days that followed realised 58.73. The window opened on 25 February 2020 priced 16.90 and realised 80.61, a span that includes the session of 23 March 2020, a fall of 12.98 per cent.

Then the gap reversed inside a month. India VIX peaked at 83.61 on 24 March 2020, and the window opened that day realised 52.50, the most positive gap in the record. The year 2020 holds both ends: the worst window and, at +6.54, the highest median gap of any calendar year, because the index stayed priced for a crash long after the crash had passed. How the market wide halts and the margin cycle ran through that month is measured in the March 2020 study.

Remove the 51 windows opened between 20 January and 31 March 2020, 1.7 per cent of the record, and the skewness goes from -4.69 to -0.57 while the mean rises from +2.70 to +3.07. One quarter is most of the tail, which makes every summary statistic here fragile in the direction that matters.

Two things in that table cut against the usual story. The gap is paid on movement, not on falls: the window opened on 13 September 2019 contains a session of +5.32 per cent, and the window opened on 8 May 2024 contains a fall of 5.93 per cent yet ended 4.4 per cent higher. A seller of variance loses to either direction. And the list is short: twelve episodes in a little over twelve years, 162 windows in all, one episode supplying most of the damage. The tail is not a steady tax. It is rare and lumpy, which is exactly what makes it expensive to insure.

Why the gap exists: the seller is paid to carry the crash

An index option is insurance, and insurance is priced above its expected payout for a reason that has nothing to do with inefficiency. The payout of an index put arrives after the index has fallen, which is when every other holding in the buyer's book has fallen too. Protection that pays in that state is worth more than its average payout, and the buyer pays for the difference. The seller commits capital to exactly the states in which capital is scarcest. Work on the major US index options found the same premium, strongly negative on average when measured as realised variance minus a variance swap rate synthesised from options, and found that the market's own beta explained only a small part of it (Carr and Wu, Review of Financial Studies, 2009). The premium prices variance risk itself.

India's positions data shows who pays and who is paid. The exchange publishes open interest by participant category every session, counting index option contracts long and short for foreign institutions, domestic institutions, members' proprietary books, and clients, a category holding every other account, individuals among them.

The client category was net short index puts on 96.6 per cent of sessions, and on every one of the 27 sessions from 20 February to 31 March 2020. Foreign institutions were net long on 99.3 per cent and domestic institutions were never net short. Proprietary books were net writers on 99 per cent of sessions in 2016 and on 13 per cent in 2026. Read these with care: they are contract counts, blind to strike, expiry and premium, so a far out of the money put counts the same as one at the money, and they cover one exchange at the close. They show where the insurance is written, not which accounts profit from writing it.

The premium is visible in data only because the event it pays for is rare. In 2017 the index sat above the volatility that followed in all 248 windows, the narrowest gap being +0.39 points. In 2023 it was above in 94.7 per cent, with a worst gap of -1.15. A participant whose record began in either year held evidence that the premium never fails. The premium is the price of the 2020 quarter spread across every other month, and a sample that happens not to contain such a quarter describes a lunch that looks free.

Harvesting it: a capped payoff, margin that rises into the loss, and a heavier cost stack

The measured gap is gross, frictionless and taken on a number nobody can trade. The exchange's full derivatives file for 18 September 2026 lists futures and options on six indices and no contract of any kind on India VIX, so the premium can only be approached by selling options, and each step from the published gap to a position subtracts something from it.

The shape comes first. A short variance position keeps at most the variance priced, when nothing at all is realised, and has no floor, because the loss grows with the square of what arrives. Option positions carry the same shape less cleanly. A delta hedged short option earns the difference between implied and realised variance weighted by its gamma along the path, so the same month can pay or cost depending on where the index sits against the strike when the moves arrive, which is the gamma cost of short convexity. The accrual that funds it is theta, which is not a daily rent, and a book spread across expiries carries vega that does not add across tenors. An unhedged short straddle simply pays away the size of the move.

The instrument is not the index either. A seller of at the money options sells at the money implied volatility, which sat a median 0.98 points below the published index. On the 277 rebuilt sessions whose following thirty days have closed, the median gap to realised was +3.30 against the index and +2.32 against at the money volatility, and the share of positive windows fell from 79.8 to 75.8 per cent. The extra point in the index is priced in the wings, which is where the crash pays out.

Then margin. Initial margin comes from the clearing corporation's portfolio scenarios, sized to a one day 99 per cent value at risk under NSE Clearing's published margin framework, and a value at risk rises with measured volatility. On top sits an extreme loss margin of 2 per cent of notional for index options, 3 per cent where the strike is more than 10 per cent out of the money against the previous close, 5 per cent beyond nine months, and since 20 November 2024 an additional 2 per cent on short index options on their expiry day (SEBI circular SEBI/HO/MRD/TPD-1/P/CIR/2024/132 of 1 October 2024, clause 5.6). None of it scales with the premium collected.

Then the cost stack, which changed twice in eighteen months. Securities transaction tax on the sale of an option, charged on the premium under section 99 of the Finance (No. 2) Act 2004, rose from 0.0625 to 0.10 per cent on 1 October 2024 (NSE/FATAX/63809, 9 September 2024) and to 0.15 per cent on 1 April 2026 under the Finance Act 2026 (NSE/FATAX/73524, 31 March 2026). On the sale of futures it went from 0.0125 to 0.02 and then to 0.05 per cent. Any costing of an option selling programme at the old rates is now wrong, and wrong where it hurts: the tax on the premium is small beside the premium, while the tax on the futures used to hedge it is charged on the full value sold. Selling one lot of futures to rebalance a hedge at the price of 18 September 2026 carries ₹759.85 of the tax, which is 7.8 per cent of the at the money premium, against ₹189.96 at the rate in force until September 2024. Exchange transaction charges, the regulator's turnover fee, stamp duty on the buying leg, goods and services tax on brokerage and charges, and brokerage itself all come on top.

Last, the part an average hides. Close to two fifths of the index's squared daily movement arrives overnight, when no hedge can be adjusted. The scenario margin is sized from a value at risk that rises with measured volatility, so it rises after the move and calls for cash on the side that is already losing. And the windows that realise more than they priced are fat in exactly the direction a seller holds, which is why a standard deviation is the wrong ruler for them, as the tail risk study measures on the same index.

What the measurement is for: describing how rich options are, not timing their sale

The comparison that can be made on the day uses the backward window: India VIX divided by the volatility realised over the previous thirty calendar days, on the same clock. Across the record that ratio has a median of 1.25, a middle half from 1.09 to 1.43, and 5th and 95th percentiles of 0.87 and 1.79; the index sat below trailing realised on 13.5 per cent of days. On 18 September 2026 it read 11.39 against 7.91, a ratio of 1.44 at the 77th percentile: options were priced richer against the past month's movement than on about three days in four.

That is a description, and the record says it is only a description. The rank correlation between the backward spread on a day and the gap that followed is 0.12 across 3,027 overlapping windows, which leaves almost all of the outcome unexplained. The worst window in the record opened at a ratio of 1.02, the 16th percentile, and two sessions earlier the ratio was 0.88, the 5th: options looked cheap against the recent past just before the most expensive month in the record to have sold them. On 31 May 2024 the ratio was 2.32, the 99.3rd percentile, and the thirty days that followed still realised 28.26 against 24.60 priced.

Used as a description it earns its place. It says what protection costs against the movement just seen, which is what a hedger pays and a writer is paid, and it sets a volatility quote beside a number in the same units. The level of the index read on its own is a separate question with its own traps, taken up in the India VIX zone study. Reading a volatility price against a stated estimator, with the window, the clock and the unit named, is a matter of method and judgement rather than a level to memorise.