A volatility index is not a forecast and it is not an average of implied volatilities. It is a model-free calculation: from a strip of out-of-the-money puts and calls at a single expiry, the market's expected variance over the period is recovered as a weighted sum of option prices, with each strike weighted by ΔK / K², and the result is annualised and expressed as a volatility. Two expiries are interpolated to hold the horizon at a constant 30 days. The methodology was published by the Chicago Board Options Exchange and is used, under licence and with local option chains, for the Indian equivalent. The number has no model of the underlying's distribution inside it, which is the whole point of the construction and the reason it can be replicated.
Why do long-volatility products lose money?
You cannot buy the index. What trades is a futures contract on its future level, and the term structure of those futures is normally in contango: the market pays more for volatility further out, because a spike is more likely to have happened by then and because the natural buyers of protection are further out the curve. That shape is what makes the obvious instrument, an exchange-traded product holding a constant-maturity long volatility position, a losing hold.
Advertisement
The mechanism is a daily roll. To hold a constant 30-day maturity, the product sells a fraction of its front-month position each day and buys the second month, which means selling the cheaper contract and buying the dearer one every single session. With the spot index at 15.00, a front future at 16.575 and a second at 18.150 on a term structure that never moves, the product loses 0.45249% a day, 9.08% over one 21-day roll cycle, and 68.11% over twelve cycles. The volatility index ends the year at exactly 15.00, unchanged, and the product that tracks it is down more than two thirds.
That figure is the answer to the question people actually ask. To be flat over a year of that decay, the constant-maturity futures level has to rise by a factor of 3.14, which is a volatility event of a scale that occurs a few times a decade. A long-volatility exchange-traded product is therefore a position with a known, computable, large negative carry and a payoff concentrated in rare events, which makes it a defensible tactical instrument held for weeks and an indefensible strategic one held for years. The arithmetic is identical to a commodity index's negative roll return in contango, which the alternatives branch derives from the same curve shape, and reading the two together is worth an hour.
What is a variance swap?
The clean instrument for the same exposure is a variance swap, which pays the difference between realised and strike variance on a variance notional, with no roll and no term-structure drag. Its awkwardness is that it settles in variance, not volatility, and variance is convex in volatility. On a $250,000 vega notional struck at 18.00, the variance notional is $6,944.44 per variance point. If realised volatility comes in at 15.50 the payoff is −$581,597.22, where a naive linear estimate of vega times the volatility difference would say −$625,000; the convexity is worth $43,402.78 to the buyer on the downside. If realised comes in at 22.00 the payoff is +$1,111,111.11 against a linear estimate of $1,000,000, and the convexity is worth $111,111.11 the other way. A long variance position gains more than linearly on a spike and loses less than linearly on calm, which is exactly the asymmetry a volatility buyer wants and exactly the reason the strike is set above the market's own forecast.
Volatility is buyable. What is cheap to buy and what is cheap to hold are separate questions.
{
"skill": "Building and stress-testing a margin and overlay model: a SPAN-style scenario array, a contract-count calculation, and a roll-decay path",
"use": "Have an assistant write the code for the sixteen-scenario risk array (pricing each leg under every price and volatility shock), for the contract-count and effective-parameter arithmetic of an overlay, and for a constant-maturity roll path, then have it write a second, structurally different implementation of the same calculation so the two can be diffed before either is believed.",
"tools": ["Chat assistants with code execution - Claude, ChatGPT, Gemini - running Python (numpy, scipy.stats.norm) rather than predicting arithmetic", "A spreadsheet holding the same scenario grid, built independently, as the diff target", "The clearing corporation's own published margin file or margin calculator, as the external check"],
"prompt": "Position: <LEGS with instrument, side, quantity, strike, expiry, lot/multiplier>. Market: <UNDERLYING PRICE, IMPLIED VOLATILITY, RATE, DAYS TO EXPIRY>. Parameters: price scan range <X%>, volatility scan range <Y vol points>, extreme move <2x> covered at <35%>. Build the full sixteen-scenario array with one row per scenario showing the shocked underlying, the shocked volatility, each leg's revalued price and P&L, and the net. Report the worst cell as the scanning risk. Then add exposure margin at <Z%> of notional and subtract the net option value. Show every intermediate at full precision and state which option pricing convention you used and why.",
"verify": "Recompute at least three scenarios by hand, including the worst cell and one extreme scenario, and confirm the 35% coverage was applied to the extreme rows only. Check the option pricing convention: a contract on a future is priced off the FORWARD, and using a spot-based formula silently misprices every cell. Then compare the total against the clearing corporation's own margin calculator for the same position and reconcile any gap to a named parameter rather than accepting it. Trace the price scan range, the volatility scan range and the exposure rate to the current margin circular with the date recorded (Guardrail 1 and 4), because all three are revised and a stale scan range produces a confidently wrong requirement. A margin number you cannot reconcile to the clearing corporation's own file is a hypothesis about your own solvency.",
"diy": "You must be able to build the array by hand for a two-leg position, identify which scenario will bind before you compute it, and convert the result into both a gross-notional and a delta-equivalent leverage figure. Worked Example 3 is the drill. The model can generate sixteen rows in a second; it cannot tell you that a covered call's binding scenario is the full down move rather than the extreme one, and that judgement is what stops you from accepting a wrong array.",
"market": "IN"
}