Guide · Derivatives and risk · Updated 2026-09-19

Delta is a hedge ratio that happens to sit near a probability, and the two are never the same number

Every number below is computed from the inputs stated in each caption, and the three formulas involved are checked against a brute force search, an exact symmetry and an independent simulation before the page is written. The argument here is a measured gap, and an asserted one would be worth nothing.

Two expressions, not one quantity with two names

Delta is the rate of change of the option value with respect to the underlying price. It exists because somebody wanted to know how much of the underlying to hold to be insensitive to a small move, and it answers that exactly. It is a hedge ratio. Nothing in its definition mentions probability, and it would still be correct in a world where nobody had a view on where the underlying was going.

What happens next is an accident of algebra. Work the derivative through for a call under the standard assumptions and it comes out as N(d1), the normal distribution function at

d1 = [ ln(S / K) + (r + 0.5 v²) T ] / ( v sqrt(T) )

where S is the underlying, K the strike, r the interest rate, v the annual volatility and T the time left. Separately, ask the model what the chance is that the underlying finishes beyond the strike, and the answer comes out as N(d2), where

d2 = d1 minus v sqrt(T)

That is the whole mechanism. Two different questions produce the same function at two points separated by the total volatility to expiry. When that separation is small the numbers are close, and the closeness is what the shortcut lives on. When it is not, no care in choosing the volatility input will make them agree, because the disagreement is not an estimation error.

One consequence follows at once. Since d1 exceeds d2 while an option is alive, a call delta always sits above the corresponding probability, and the size of a put delta always sits below it by exactly the same amount. A book reading delta as probability on both legs overstates one and understates the other, which is the kind of error that cancels in a net figure and survives in every gross one.

The gap, measured

The table below fixes a volatility of 15 per cent a year, 30 days to expiry and a rate of 5.50 per cent, and walks the strike across the range in which most options trade. The current level is normalised to 100, so a strike reads as a percentage of it.

Two readings point in opposite directions and both are correct. In absolute terms the gap peaks in the middle, at 1.72 points around the money, and falls away on either side. In proportional terms it does the reverse: at a strike of 105 the gap is 6.9 per cent of the probability, at 115 it is 15.8 per cent, and there is no bound on it. Both quantities go to zero out in the tail, and the delta gets there more slowly.

The second reading is the one that costs money, because the tail is where the shortcut gets used to justify a position. A strike picked on the reasoning that the delta reads 0.02 and the event is therefore a two in a hundred affair is being picked on a number more than a tenth larger than the model probability it is standing in for, before anything else has gone wrong.

One number governs how wrong the shortcut can be

The maximum of the gap over all strikes has a closed form, and it is unusually clean. Setting the two arguments symmetric about zero puts the maximising strike at the forward price, and gives the maximum gap as

2 N( v sqrt(T) / 2 ) minus 1

The strike does not appear. Nor does the current level. Nor does the interest rate, which the build script verifies by re-running the search at rates from zero to 12 per cent and finding the same peak every time. Only the total volatility to expiry survives, and that single number sets the worst the shortcut can do.

Read the top left corner and the shortcut looks defensible: a quiet weekly option cannot be more than about half a point out, and half a point sits inside the width of a spread. Read the bottom right and it collapses. A year out at 45 per cent volatility the two numbers can be seventeen points apart, which is not a refinement, it is a different answer.

What the table implies about habit is the uncomfortable part. A trader who learns the shortcut on weekly contracts, where it works, carries it unchanged onto longer dated positions, where it does not, and nothing in the experience marks the transition. The formula does not stop applying. It stops being close.

Finishing beyond a strike is not the same as reaching it

The larger error lives here. Almost nobody who says a strike has a twenty per cent chance means the underlying will be beyond it at one specific instant several weeks away. They mean something about the position being threatened, tested, adjusted or assigned. Those are statements about the whole path, and a path can cross a level and come back.

The probability of reaching a level at any time before expiry follows from a reflection argument applied to the log of the price. With no drift in the log it is exactly double the probability of finishing beyond that level. With drift it is close to double, and the computed values say how close.

The first row is the sharpest. For a strike sitting at the current level the delta reads 0.55 and the probability of finishing beyond reads 0.53, so the shortcut delivers something in the region of a coin toss. The probability of touching that strike is 1. The level is already there. It is not a matter of chance at all, and the two numbers people argue over are both answering a question nobody asked.

Further out the ratio settles between 1.89 and 1.94, rising toward two as the strike moves away and the drift matters less. The translation is blunt: roughly double the delta before treating it as the chance of a level being reached. That correction is an order of magnitude larger than the gap this page opened with, and it runs in the direction that makes a position look worse.

The probability inside a price is a price

Suppose all of that is handled: the right expression is used, and the path question is kept apart from the endpoint question. There is still a problem with calling the result a probability, and it is not a technicality.

An option price is not built by forecasting. It is built by asking what it costs to manufacture the payoff by trading the underlying. Prices set that way can be written as an expected payoff under a particular set of probabilities, and that set is not the one a forecaster would write down. It is tilted, so outcomes that are painful to hedge carry more weight than frequency alone would give them, and the tilt is what makes the expected payoff equal the cost of producing it.

What comes out is a price wearing the clothes of a probability. It says what the payoff is worth today, not how often the event happens, and the two can differ persistently without either being wrong. The size of the effect is easy to show. The price implied figure uses the interest rate as the drift, because the hedging argument forces that. Substitute any other drift and the probability moves.

The spread across those rows is comparable to, and in places larger than, the delta to probability gap measured in the first half of this page. Unlike that gap, this one cannot be computed away. The drift is not in the price. Anyone who wants a real world probability has to supply a view on it and own it as an assumption.

Delta stopped being only an analytical quantity in 2025

All of this would be intellectual hygiene if delta stayed on the analysis side of the desk. In Indian equity index derivatives it no longer does.

Through 2025 the market regulator moved position limit measurement in equity index derivatives from a contract count onto a delta adjusted futures equivalent basis, in a framework for intraday position limits monitoring issued in circulars dated 29 May 2025 and 1 September 2025. The exchange standard operating procedure implementing it, dated 30 September 2025 and applicable from 1 October 2025, does something unusual: it does not merely require a delta, it prescribes the formula. In its own terms the futures equivalent of a call is N(d1), of a put is N(d1) minus 1, of a futures contract is 1, and d1 is the Black and Scholes expression given above.

That is the same N(d1) discussed above, written into a rulebook. It sets the intraday net limit per entity per index at Rs 5,000 crore on a futures equivalent basis, against an end of day limit of Rs 1,500 crore, and an intraday gross limit of Rs 10,000 crore separately on each side. Compliance is checked through at least four random snapshots during the day, including one between 14:45 and 15:30, with a fifteen minute cure period after a breach that does not apply to that final snapshot.

The operationally important detail is what goes into the formula, because the prescribed inputs are not the ones a pricing screen uses.

The volatility row is the one that bites. Take a single call line, at a strike two per cent below the current level with seven days left, whose quantity would be worth Rs 6,000 crore if every unit counted at a delta of one. The prescribed delta on that line, and so its futures equivalent value, depends entirely on which volatility figure goes in.

The position never changed. Only the volatility input did, and the same line sits on either side of a limit depending on which figure went in. An entity computing its own delta from the volatility the option market is quoting, rather than the previous day close figure the procedure prescribes, arrives at a different futures equivalent number for a book it already holds, and the limit is measured on the prescribed one. The absent dividend term is smaller but real: at an assumed index yield of 1.5 per cent the prescribed delta on an at the money strike sits about 1.20 points from a dividend adjusted delta at 30 days, and 4.64 points at a year.

Two further dates matter. The transitional arrangement under which participants leaned on exchange published figures ran until 5 December 2025 and ceased with effect from 8 December 2025, from which point entities intending to take large delta based positions in index options are expected to run their own delta computation. From the same date an intraday breach on an options expiry day additionally attracts an additional surveillance deposit, levied at one and a half times the computed amount and blocked for a month.

So the sequence is complete. A quantity many traders carry around as a rough probability is now the quantity a limit binds on, computed by a published formula from specified inputs that are not the ones on a screen, checked at unannounced moments, with a deposit consequence on expiry day. Anyone sizing near these limits should also read how market wide position limits and the ban period interact with it, since the two bind on different measures of the same book.

What delta is reliably for

None of this argues against delta. It argues against one use of it. Delta does two jobs extremely well, and both are the jobs it was derived for.

The first is hedging. Delta is the quantity of the underlying that neutralises the position against a small move, and by construction it is exactly right for that. It degrades as the move gets larger, which is what the second derivative describes, but within its range it is not an approximation to something else. It is the answer.

The second is aggregation. Contract counts across different strikes and expiries cannot be added, because a contract far from the money and one at the money are not comparable quantities of exposure. Delta adjusted quantities can be, and the sum means something: the equivalent position in the underlying. That is why the regulator chose it for limit measurement, and it is a sound choice. A limit on contract counts constrains nothing, since the same exposure can be assembled from a very different number of contracts.

The failure mode in both uses is quieter than the probability confusion. Delta is local. It describes a small move from where the underlying is now, under the volatility and time assumed now. A book flat on delta is flat only in that neighbourhood, and the neighbourhood shrinks as expiry approaches. Reading a delta adjusted total as a measure of risk rather than of exposure at one point is a separate mistake, and at least as common. The same discipline applies as in managing the risk of a short option book: the number that aggregates is not automatically the number that warns.

The order in which to ask

The repair is a sequence, not a formula. Decide first which question is on the table, because three quantities have been collapsed into one word. For the endpoint, use the expression with the smaller argument. For whether a level gets reached at all, the answer is a path quantity, close to twice that, and it is the one most people meant. For the real world rather than the price, a drift assumption has to come from outside the option market and be owned as an assumption. And for how much underlying to hold, or how much exposure a book carries, delta is simply the right number. It stops being reliable only when a hedge ratio is asked to do the work of a forecast, and the failure is silent, because the number it returns is always plausible.