Delta Hedging: What It Costs to Stay Neutral, and Why That Rules Most People Out

The mechanism is the easy half

An option's delta is the rate at which its price responds to a one point move in the underlying. A call struck at the money has a delta near one half, meaning it gains about fifty paise for every rupee the index adds. A writer of that call is therefore short half a unit of index exposure for every unit the contract covers, and can cancel the exposure by holding half a unit long. Do that and the pair is, for a moment, indifferent to direction.

The words "for a moment" carry the entire subject. Delta is not a constant. It rises toward one as the option moves into the money and falls toward zero as it moves out, and the speed of that change is gamma. A written option is short gamma, which means the required hedge always moves in the wrong direction: as the index rises you have to buy more, at a higher price, and as it falls you have to sell, at a lower one. Every adjustment locks in a small loss by construction. Our sibling page on risk management for option writers describes that seller-side Greek profile in words; this page puts a price on it.

That is also why delta hedging is a different animal from the hedges most retail material describes. A protective put or a static short futures position against a cash portfolio is set once and left, and our guide to what a hedge actually is covers those four static structures in detail. A delta hedge is not a position, it is a maintenance schedule. The interesting question was never how to compute delta. It is how much the schedule costs and who can afford to run it.

There is a second reason the mechanism gets over-explained and the economics under-explained. The mechanism is the same everywhere in the world. The economics are local: they depend on the tax on a futures sale, the size of one contract, and the tick you cross to get filled, and all three of those are Indian facts with dates attached to them.

The position, stated in full so it can be checked

Everything below comes from one configuration. It is a synthetic index rather than a live instrument, so that the regimes are stated rather than inferred and so that nothing here implies a specific security would have behaved a specific way. Every parameter is listed, including the ones that are assumptions rather than verified facts, because the difference between those two categories is the difference between a computation and a guess.

Two entries in that table deserve emphasis. The first is the lot. Indian index derivatives are sized to a rupee value, not a share count: the exchange solves the lot backwards so that one contract sits inside a minimum contract value band, which SEBI raised to 15 lakh to 20 lakh rupees with effect from 20 November 2024. The lot is therefore a control output, not a constant. It went from 25 to 75 for the headline index at that revision, and was trimmed again from the January 2026 contracts as the index appreciated. The 65 units used here is an illustrative figure chosen so that the contract value lands inside the verified band at the stated index level, and it should not be read as the present lot for anything. Our page on how a lot size is decided carries the dated table and explains the mechanism that keeps moving it.

The second is the interest rate, which is set to zero along with the dividend yield. That makes the index future trade at the index level and makes the futures hedge ratio equal the option delta exactly. It is a simplification, and it removes carry from the exercise. Carry is a small one off amount here and it would shift every number on this page by the same direction and roughly the same size, so removing it does not change the shape of a single curve. What it buys is that every rupee below is friction, not financing.

Realised volatility is set equal to implied volatility. That is the most important choice in the whole setup and it deserves to be stated as a choice. It means the option is priced fairly: the writer has no edge, and no disadvantage, before costs. The consequence is that the average outcome of a perfectly hedged writer is exactly zero minus the friction, so every rupee of average loss reported below is a cost and nothing else. If you assume implied volatility exceeds what the market will realise, you are assuming the answer.

What one adjustment costs

A rebalance is a futures order, and a futures order in India pays a stack of charges that are mostly not negotiable. The site's verified charge research is the source for the statutory lines. Securities transaction tax on the sale of a futures contract stands at 0.05 percent of contract value from 1 April 2026, under section 159 of the enacted Finance Act 2026 amending serial 4 of the section 98 table. Stamp duty on the buy side is 0.002 percent under Article 56A(d) of Schedule I to the Indian Stamp Act 1899. Goods and services tax at 18 percent applies to brokerage, exchange charges and the regulator's turnover fee, and does not apply to transaction tax or stamp duty.

Put those against the illustrative contract value of 15,60,000 rupees and one round trip in one futures lot comes to about 1,001 rupees. Of that, 780 rupees is transaction tax on the sale leg alone. Roughly seventy eight paise in every rupee of rebalancing cost is a statutory levy on one side of the trade, which has three consequences worth holding on to.

The first is that the cost is asymmetric. Buying a lot costs about 126 rupees and selling the same lot costs about 875 rupees. A hedging programme that ends up net long futures at expiry and closes out pays that asymmetry once more at the end. The second is that the cost is proportional to contract value, not to how far delta moved, so a rebalance that adjusts the hedge by a tenth of a lot and one that adjusts it by a whole lot are not remotely comparable in efficiency. The third, and the one that matters most for the rest of this page, is that the dominant term is the one you cannot shop around for. Brokerage is commercial and varies; the exchange charge is a published rate; transaction tax is legislation.

Against a premium of 25,150 rupees for the written lot, a single round trip is 3.98 percent. Do this thirty times over the option's life and the arithmetic starts to speak for itself before any simulation is run. The simulation exists to answer the question that arithmetic alone cannot: how many times you actually have to do it.

The trade-off, and where the total stops falling

Hedge rarely and the position drifts away from neutral between adjustments, so the outcome is dispersed. Hedge constantly and the outcome is tight but the charge stack eats the premium. Both statements are obvious. What is not obvious is where the two curves cross, and that is a computation, not an opinion.

The measure used here is the root mean square of the writer's final outcome across all paths, which is simply the square root of the average squared result. It needs no arbitrary risk aversion parameter, because it penalises a large average cost and a wide spread on the same footing: hedge too little and the variance term dominates, hedge too much and the mean term does. The minimum of that curve is a defined point.

Two families of rule were tested. A fixed-clock rule rebalances every so many minutes regardless of what the index has done. A delta-band rule rebalances only when the hedge has drifted more than a set amount away from the target, however long that takes. The band rule wins, and it wins on both axes at once: at its optimum it used 29.6 adjustments against the clock rule's 43.5, and still left less risk. That is not a subtle effect and it has a plain explanation. A clock rule spends money on adjustments the market did not ask for, and misses the ones it did.

Read the table from the top and the shape is unmistakable. Setting the hedge once and never touching it costs almost nothing and leaves 76.1 percent of the premium at risk. Rebalancing once a day cuts the leftover risk to 18.7 percent of the premium for a cost of 6.9 percent. Push to the optimum and the total lands at 14.4 percent. Push past it and every further adjustment buys a smaller reduction in risk than it adds in cost, until at one adjustment every five minutes the writer has spent 171.1 percent of the premium and finished materially worse off than if the option had simply been left alone.

That last row deserves to be sat with. Hedging more is not conservative. Past the optimum, more hedging is a strictly worse position on both dimensions of the trade-off at once, because the transaction cost itself becomes a source of variance: the paths that move most trigger the most adjustments, so the bill arrives largest exactly where the writer can least afford it. On the tightest rule in the table, the 0.02 delta band, the standard deviation of the final outcome was 4,310 rupees against a residual hedging error of 810, and almost the whole difference is the variability of the charges rather than of the market.

The honest summary of this section is that delta hedging works. Run at its optimum and at a size where it can be run at all, it removed 93.5 percent of the dispersion in the writer's outcome. The problem is not that the technique fails. The problem is the two words at the end of that sentence.

What the hedge does to an actual outcome

Averages across twenty thousand paths are the right way to choose a rule and the wrong way to understand one. Two individual paths, both with rebalance counts near the middle of the simulated distribution so that neither is an outlier, show what the writer actually experiences.

On the first path the index climbs and finishes above the strike. The unhedged writer ends 55,314 rupees down on a lot that collected 25,150 rupees of premium, which is the ordinary arithmetic of writing an option that goes wrong. The hedged writer ends 2,662 rupees down. On the second path the index wanders for three weeks and comes back to where it started. The unhedged writer keeps essentially the whole premium, 25,088 rupees. The hedged writer ends 1,161 rupees down, having spent 2,477 rupees on twenty eight adjustments to protect against a move that never came.

Neither outcome is a success. That is the point, and it is the part that gets lost when delta hedging is presented as a solution. The hedge did not make the writer money on either path. It replaced a large loss and a large gain with two small losses of almost the same size. The technique is a compression of the outcome distribution around a slightly negative mean, and the mean is negative by exactly the amount of the friction. Anyone running it is not trading direction any more; they are trading the gap between the volatility they sold and the volatility the market delivered, and the friction is the toll on that trade.

The right unit for that toll is not rupees, it is volatility points, because that is the unit the option was priced in. One written lot here carries about 1,796 rupees of vega per volatility point. Inside a book large enough to hedge properly the friction came to about 1.1 volatility points, so an option sold at an implied 14 needs the market to realise under about 12.9 for the exercise to break even. That is a demanding but not unreasonable condition, and it is roughly what an options desk means when it talks about a variance risk premium. Our page on implied volatility and the gap between implied and realised covers where that premium comes from.

Why nearly all the work arrives at the end

The rebalancing requirement is not spread evenly across the option's life. It is concentrated, violently, into the last few sessions, and the reason is a clean piece of arithmetic rather than a market observation.

Delta moves at a rate set by gamma multiplied by the volatility of the underlying. For an at-the-money option, gamma is inversely proportional to the square root of the time remaining, so the speed at which delta wanders is inversely proportional to that same square root. A band rule fires when delta has wandered a fixed distance, and the expected time to cover a fixed distance under a diffusion falls with the square of the speed. Combine the two and the required rebalancing rate is inversely proportional to the time remaining. Halve the time to expiry and you double the work.

The consequences are stark. At 21 days out, an at-the-money position on a 0.05 delta band needs about three adjustments a day. At five days it needs about thirteen. On the final day it needs about 64, and in the closing hours of that day the rate passes 250 a day. Nothing about the position has changed except the calendar.

Those are the analytic rates for a position that stays exactly at the money. The simulated paths give the lived version, which is lower on average and far more concentrated. Averaged over all paths the observed rate rose only from 2.2 adjustments a day at the start to 4.8 on the final day, for the simple reason that most paths have wandered away from the strike by then and an option that is far from its strike has almost no gamma and needs almost no attention. Split the final day by where the index actually was, though, and the average dissolves into two completely different regimes: a position still within a quarter of a percent of the strike needed 22.5 adjustments in that one session, while a position more than two percent away needed 0.1.

This is the quantitative form of a warning that appears qualitatively all over options writing. Short-dated near-the-money options are not slightly harder to manage than others, they are harder by two orders of magnitude, and the difficulty appears suddenly rather than gradually. It is also why the difficulty is systematically underestimated: a writer who has managed several positions that drifted away from the strike has genuinely experienced a light workload, and has learned nothing about the one that does not drift away. The page on what actually happens on expiry day covers the market structure side of the same problem.

The lot you cannot cut in half

Everything up to here would be true in any market. What follows is specifically Indian, and it is the constraint that decides who can use this technique at all.

Delta hedging assumes the offsetting position can be set to whatever fraction the model asks for. In index derivatives you can only trade whole lots. The required hedge for one written option lot is delta lots of futures, and delta is a number between zero and one, so the achievable hedge for one written lot is either nothing or everything. There is no third setting. The instruction "reduce the hedge by 0.08 of a lot" has no executable form.

The left panel counts two different things that are usually assumed to be the same. The coral bars are the number of times the delta-band rule said an adjustment was required. The green bars are the number of times an order could actually be sent, because the rounded target had changed. At one written lot the rule fired 1,194 times over the option's life and the hedge could move 26 times. The position spent the overwhelming majority of the option's life outside its own tolerance band with no legal way back inside it.

The right panel shows what that does to the outcome. With whole-lot rounding at one written lot, the risk left over was 124.6 percent of the premium taken in, against 14.6 percent for the same rule with fractional hedging. And the cost of achieving that was 13,110 rupees, or 52.1 percent of the premium, because the 26 movements that were possible were not gentle adjustments. They were the hedge flipping between zero and one whole lot as the index crossed the strike, which is a stop and reverse programme wearing a hedging label, and it whipsaws for exactly the same reason a stop and reverse programme does.

The threshold is computable. The optimal band was 0.08 of a delta, so a book of N written lots moves its ideal hedge by 0.08 N lots per breach. For the rounded target to change, that step has to be worth about half a lot, which first happens at seven written lots. The simulation agrees with the arithmetic: at eight lots the penalty from rounding falls to 6.7 percent of the ideal, and from thirteen lots upward it turns slightly negative, because at large size the rounding acts as a free extra filter that suppresses adjustments too small to be worth their own charges.

There is a residue that never goes away, and it is worth naming because it is counter-intuitive. Rounding to the nearest whole lot leaves an unhedged sliver of index exposure, and across every book size from eight lots upward the simulation put the typical size of that sliver at between 4.0 and 4.4 lakh rupees of index exposure. It does not shrink as the book grows, because it is a property of the lot, not of the position. For a one lot writer that residue runs to about thirty percent of the entire contract. For an eighty lot book it is a rounding error in the literal sense. Large books are not better at hedging because they are cleverer. They are better at it because a fixed quantum of imprecision is small relative to them.

Where the technique becomes available, and where it does not

Putting the cost curve and the granularity constraint together produces a size threshold rather than a technique. Below it, delta hedging is not a harder version of the same thing; it is a different and worse thing that happens to share a name.

The pattern across those rows is not gradual. Between one lot and five lots the position is not being hedged in any meaningful sense, and it pays between a quarter and a half of the premium for the privilege. At eight lots the constraint releases almost all at once, and from there upward the numbers stabilise: around seven percent of the premium in cost, around twelve percent left at risk, and around thirty orders over the option's life. The threshold at the stated parameters sits somewhere just above 1.2 crore rupees of written index notional. Converting that to an account size requires a margin assumption, and the illustrative fifteen percent of notional used here puts it near 19 lakh rupees of margin, before the margin on the futures hedge itself and before any buffer for the fact that a writer's margin expands against them as the position moves. Real margin is recomputed daily by the clearing corporation and is not a fixed share of notional, so treat that conversion as a scale, not a number.

The durable part of that finding is its shape rather than its value. The lot moves, the index level moves, and the margin percentage is not a constant. What does not move is the structure: the minimum viable size for a dynamically hedged option book is set by the ratio between one futures lot and the delta step of the position, and that ratio puts the answer in crores of notional rather than lakhs. It is worth stating plainly that this is the same wall the lot size imposes on ordinary position sizing, described on our page about the lot-size floor, arriving one level higher up: there the lot decides whether you can take the position at all, here it decides whether you can manage it once taken.

It is also worth stating what this does not mean. It does not mean the option cannot be written. It means the specific risk control being described cannot be applied to it, and a writer who believes otherwise is carrying a directional position while thinking they are flat. That misapprehension is more dangerous than an openly directional position, because it is sized as though it were safe.

The alternatives, priced against it

If the technique is unavailable below a certain size, the useful question is what the same account can do instead. Three comparisons are computable on the identical set of paths, with every charge deducted and with all four positions priced at the same implied volatility so that none of them has an edge over the others. There is also a zeroth option, which is not writing the contract at all: no premium, no charges, no maintenance and no tail. It scores zero on every column below, and it is the baseline the other rows have to be measured against, because a position that is never taken cannot be mismanaged.

Writing the call and leaving it alone is the honest baseline. It costs 62 rupees, its average outcome is a loss of 486 rupees, which is zero to within the sampling noise of twenty thousand paths, and it has a very long left tail: the worst one path in a hundred lost 128,232 rupees against a premium of 25,150, and the single worst path in the run lost 264,291. That distribution is what the option premium is compensation for, and it is also why about 93% of individual traders in equity derivatives made net losses over FY22 to FY24, with aggregate net losses exceeding Rs 1.8 lakh crore (SEBI, September 2024) is the correct frame for any page on this subject.

Delta hedging that single lot with whole futures contracts moves the average outcome from roughly zero to a loss of 12,877 rupees, and shortens the left tail from 128,232 to 97,211 at the same percentile. The writer paid 13,046 rupees, more than half the premium, to remove about a quarter of the tail. Inside a twenty lot book the same rule behaves as it is supposed to: the outcome per lot has a standard deviation of 2,408 rupees against 38,246 unhedged, and the worst one in a hundred is 7,989 rupees per lot.

The fourth row is the one that makes the small-account case awkward. Buying a further out of the money call against the written one converts the position into a defined-risk spread. It gives up a large share of the premium, taking in 12,751 rupees instead of 25,150, and in exchange the maximum loss is capped by construction at 19,749 rupees. It requires two orders and no maintenance at all, and the total charge was 105 rupees. On the identical paths its outcome distribution was narrower than the whole-lot delta hedge managed, its tail is bounded rather than merely shortened, and it cost about one hundred and twenty fourth as much to run.

These are not equivalent positions and it would be wrong to present them as substitutes in general. A spread is short a different exposure than a hedged short call, and it caps the writer's participation as well as their loss. But for the specific problem of a small account trying to stop a written option from running away, the comparison is one sided, and it is one sided because of arithmetic rather than preference. There is a structure whose risk control is built into the position at zero maintenance, and a structure whose risk control has to be bought thirty times a month in units you cannot subdivide. Our framework page on defining maximum loss before entry works through how that choice is normally sequenced.

What the computation is actually for

It would be easy to finish here with the conclusion that delta hedging is for institutions and stop. That is true but incomplete, and the incomplete version loses the transferable part.

The transferable part is the habit of converting a technique into a cost before adopting it. Almost everything that is presented to a retail audience as risk management has a price, and the price is almost never stated in the same place as the description. Delta hedging is an unusually clean example because its price is computable to the rupee: a known charge stack, a known number of adjustments, a known residual. Most techniques are not that tractable, but the question is the same one. What does running this cost, as a share of what the position was going to pay me, and what does it leave behind after I have paid it?

The second transferable part is the size test. A great deal of trading knowledge is written by and for participants operating at a size where certain frictions round to nothing, and it is repeated downward without the size qualification travelling with it. Whole-lot granularity is invisible above a crore of notional and decisive below it. Ask of any technique what the smallest position is at which it still behaves as described, and a surprising amount of standard advice turns out to have a floor nobody mentioned.

The third is that a hedge which cannot be adjusted is not a conservative version of a hedge. It is a different position with a different risk, and it is more dangerous than no hedge at all, because it is sized by someone who believes they are flat. The specific arithmetic on this page will age, since the lot will be revised and the transaction tax rate has already moved twice in three years. The structure of the argument will not, and rebuilding it with current numbers is an afternoon's work. Doing that kind of arithmetic before adopting a technique rather than after is the habit worth taking away, and it is the substance of the method we teach.

Every figure comes from one seeded simulation written in Python with numpy only, reproducing identically on each run. Index paths are geometric Brownian motion with zero drift on a grid of 75 steps per trading day across 21 trading days, with realised volatility set equal to the 14 percent used to price the option, so the contract is fair and the writer starts with no edge. Option prices, deltas and gammas use Black Scholes with the interest rate and dividend yield both set to zero; the normal distribution function is a polynomial approximation checked against the standard library to within 1.4 parts in ten million, and delta and gamma are checked against finite differences of the price and of the delta respectively. The main curves use 20,000 paths, the position-size study 8,000, and the sensitivity sweep 8,000. Where the same quantity is reported from two different runs, for example the cost of hedging a single lot, the two figures differ only by sampling noise, which here is about half of one percent.

Statutory charges are taken from the site's verified charge research: transaction tax of 0.05 percent on a futures sale and 0.15 percent on an option sale from 1 April 2026, stamp duty of 0.002 percent on a futures purchase and 0.003 percent on an option purchase, and goods and services tax at 18 percent on brokerage and exchange charges but not on transaction tax or stamp duty. Brokerage, the exchange and regulator charge, and the half spread are commercial or segment figures rather than verified statutory rates; they are labelled as assumptions in the setup table and each was separately set to zero and doubled. The optimal rule was unchanged in every variant except the one where brokerage and exchange charges were removed entirely, and the total moved within a range of 2,856 to 3,821 rupees against 3,706 as modelled on the same 8,000 paths.

All results are illustrative and simulated. They are not a track record, not a forecast, and not an indication of what any position would produce in a live account. Nothing here is a recommendation to write, hedge or hold any instrument. The purpose is to show the relationship between rebalancing frequency, transaction cost and residual risk, and the effect of a whole-lot trading unit on that relationship, both of which are properties of the arithmetic rather than of any particular market.

• What a hedge actually is, and what each one costs →
• How a lot size is decided, and the floor it puts under an account →
• Implied volatility, and the gap between implied and realised →
• Risk management for option writers →