The Ratio Spread: How a High-Probability Structure Carries Ruin Risk

What the extra short leg actually is

Start with a structure that is not a ratio spread. Buy one call at a lower strike, sell one call at a higher strike, and you own a vertical spread. Everything about it is bounded. The most you can make is the distance between the strikes less what you paid, the most you can lose is what you paid, and both numbers are known before the order goes in. The short leg in that structure is covered, in the precise sense that the long leg you also own will pay out exactly what the short leg costs you above the higher strike. The two cancel, and the payoff line flattens.

Now sell a second call at that same higher strike and change nothing else. The first short call is still covered by the long. The second one is not. There is no third option in the position to pay for it. Above the higher strike the structure is short one uncovered call, and an uncovered call has no upper bound on what it can cost, because there is no upper bound on where an index can settle. That is the whole mechanism. Everything else on this page is arithmetic that follows from it.

Two things happen at the moment the second call is sold, and only one of them appears in the marketing. The first is that the extra premium can turn a debit into a credit. Instead of paying to open the position you are paid to open it, and the payoff line below the lower strike sits above zero rather than below it. The second is that the position's worst case stops being a number and becomes a direction. Both are true simultaneously, and the first is much easier to put in a screenshot.

It is worth being exact about the word covered, because it is doing real work. A short call is covered when something else in the account will move in the opposite direction, point for point, no matter how far the underlying travels. A long call at a lower strike does that, but only for one short call. Adding a second short call against the same long adds an exposure that nothing offsets. The position is a spread plus a naked short, and the two halves behave so differently that averaging them into one description of the structure is what causes the misunderstanding.

This page assumes you already know how a single option behaves and what a strike does. If either is unfamiliar, the options framework guide is the place to start, because none of what follows will land without it. What follows is one structure, computed end to end, on a simulated index rather than a real one, so that every number can be reproduced or disputed.

The payoff, computed

The worked example uses an illustrative index at 25,000 with thirty days to expiry, implied volatility of 15 percent held flat across strikes, and a financing rate of 6.5 percent. One standard deviation of movement over the life of the contract is 1,075 index points, which is 4.30 percent. The lot is illustrative and was chosen so that contract value sits inside the current minimum band; lot sizes are revised periodically and no number quoted here should be read as the present one.

The row that most often surprises people is the one about breakevens. A ratio spread is usually drawn with two, and a ratio spread entered for a debit does have two. This one does not, and the reason is worth stating because it is the source of the high win rate. If the account was paid to open the position, then any settlement at or below the lower strike leaves it with the credit and nothing else. There is no price on the downside at which the position turns negative, because there is no downside cost to recover. The profitable zone therefore runs from zero all the way up to 26,050.60, which is why the structure wins as often as it does.

Change the short strike from 25,500 to 25,750 and the arithmetic flips. Two calls at 25,750 are worth 194.80 each against a long at 497.60, so the structure costs 108.00 points, Rs 6,480, to open. Now there are two breakevens, at 25,108.00 and 26,392.00, the profitable zone is 5.14 percent of spot wide, and the probability of settling inside it is 37.74 percent. The peak is higher at Rs 38,520, but it is reached from a position that starts underwater and has to be carried there. That version of the structure is arithmetically honest about what it is, and it is not the version that gets described as high probability.

The figure is drawn on a single vertical scale so that the two halves can be compared, and the comparison is the point. The profit is a triangle 550.60 points tall at its apex, reached only if the index settles exactly at the short strike. The loss is a straight line falling one index point of value for every index point the settlement is above 26,050.60, and it does not stop at the edge of the frame. At 27,000, a move of 8 percent, the loss is Rs 56,964. At 27,500 it is Rs 86,964. At 30,000 it is Rs 2,36,964, and it is still falling at the same rate.

One structural detail deserves a sentence of its own. The rate at which the loss grows is set by the ratio, not by the strikes. Selling two against one leaves one uncovered call, so the loss grows one point per point. Selling three against one leaves two uncovered calls, so it grows two points per point, and the same adverse settlement costs twice as much. Widening the gap between the strikes moves the breakeven further away, which buys time and probability, but it does not change the slope. Nothing available inside the structure changes the slope except reducing the ratio, and reducing the ratio to one is the same thing as not having a ratio spread.

Winning often, and the size of the exception

A payoff diagram tells you what happens at every settlement price. It does not tell you how likely each of those settlements is, and that omission is where the structure hides. So the same structure was run through 5,00,000 simulated expiries, drawn from the distribution the option prices themselves imply, which makes it the market's own estimate rather than anybody's forecast.

The outcome split is stark and it is the reason this structure sells itself. Four expiries in five finished in profit. Another 8.57 percent lost less than Rs 25,000, which on most accounts is an ordinary bad month. Losses above Rs 75,000 accounted for 2.58 percent, about one expiry in 39, so a trader running this once a month would meet one roughly every three years and would learn to think of that as the bad case. Losses above Rs 1,50,000 were 0.14 percent, one expiry in 713, which at monthly expiries is once in about sixty years. That last number is the one to hold lightly, because it is the number the two models on this page disagree about most: the wider tailed one puts the same loss at 0.67 percent, one expiry in 150, which is once in about twelve years rather than sixty.

Now put the two halves of the distribution side by side. The average gain on a winning expiry was Rs 9,292. The average loss on a losing expiry was Rs 38,143, four times larger. Multiply each by how often it happens and the two products cancel almost exactly: the expectation across all 5,00,000 expiries was Rs 18 below zero, which is the tick rounding on the quoted premiums and nothing else. This is not a coincidence and it is not a criticism of the structure. It is the pricing identity. The options were priced under the same distribution the outcomes were drawn from, so the structure is worth exactly what it costs, and no arrangement of strikes and ratios changes that.

Which means the high win rate is not evidence of anything. It is a description of the shape of the payoff, not of its value. A structure that wins four times out of five with a zero expectation is telling you, precisely, that the fifth outcome is four times the size of the other four. Reading the win rate as a measure of quality is reading one number from a pair and discarding the other.

The second column is the part worth sitting with. It is the same structure at the same premiums, run against a distribution that puts one expiry in twenty into a state three times as wide, with the quiet state solved so that the at the money call prices identically in both worlds. The overall standard deviation rises from 15.00 percent to 16.13 percent and the excess kurtosis is 4.65, which is a moderate assumption rather than an extreme one. Under it, the win rate does not fall. It rises, by 1.29 percentage points, because a distribution with fatter tails also has more mass in the middle. The 99.9th percentile loss, meanwhile, more than doubles.

That is the whole argument on this page in one line. A change to the distribution that made the tail twice as deep made the win rate slightly better. Every statistic a trader is likely to look at moved in the reassuring direction, and the only statistic that moved in the other direction is the one that decides whether the account survives. Two disclosures belong here. The first is that at the 90th percentile the wider tailed model actually shows a smaller loss, Rs 25,354 against Rs 29,026, for the same reason the win rate rose. The second is that under the wider tailed model the two calls being sold are worth 275.40 rather than 274.10, so holding the premium fixed at 274.10 underpays the structure by 2.57 index points, Rs 154.33 an expiry. That is the entire size of the inconsistency this comparison introduces, and it is disclosed rather than smoothed over.

One adverse expiry, priced in winning months

Percentile losses are hard to feel. A more useful unit is the winning expiry, because that is the unit the account has been accumulating. At Rs 9,292 of average gain, every adverse settlement can be converted into the number of prior winning expiries it erases.

A 6 percent move, which the thin tailed model puts at 10.52 percent likely in any given expiry, costs 2.9 winning expiries. That is unpleasant and entirely survivable. An 8 percent move costs 6.2. A 10 percent move costs 9.4, which is most of a year of accumulated gains removed in one settlement. Beyond that the arithmetic stops being gradual: 15 percent costs 17.5 winning expiries and 20 percent costs 25.5, which is more than two years.

The crossing points are the numbers to carry away. One year of average gains is erased by a settlement at 27,913, a move of 11.65 percent, which is 2.71 standard deviations. Two years is erased at 29,771, a move of 19.08 percent. Three years is erased at 31,630. Under the thin tailed model those three events have probabilities of 0.69 percent, 0.0038 percent and effectively zero per expiry. Under the wider tailed model they are 1.18 percent, 0.33 percent and 0.11 percent. The first pair differ by less than a factor of two. The third differ by a factor no honest model can pin down, and the entire question of whether this structure is survivable lives inside that disagreement.

Notice what this does to the psychology of running the structure. Each individual expiry gives an overwhelming impression of working. The record accumulates, the win rate stays where the diagram said it would, and the position size drifts upward because nothing has argued against it. The loss, when it comes, does not arrive as a series of warnings. It arrives as one settlement, and its size is set by how far the index travelled that month, which is a quantity nobody controls and nobody can bound in advance.

Ruin, computed rather than asserted

Saying that a loss is unbounded is true and not very useful, because in practice the question is whether the account survives a given number of repetitions. That is a computable question, so here it is computed. Take an illustrative account of Rs 20,00,000. Every expiry, open four of the structures above. Hold to settlement, pay the charge stack, and repeat sixty times. Ruin is defined as equity falling to half the starting account or below, on the reasoning that a fifty percent drawdown ends the ability to carry the same size, which is the definition our risk of ruin page works from and which this page does not rebuild.

Four structures is not an aggressive size by the measures a trader normally applies. The best possible outcome in any expiry is Rs 1,32,144, which is 6.61 percent of the account. It is aggressive by the only measure that matters here: the uncovered notional standing behind the position is Rs 61,20,000, which is 3.06 times the account. That number is available before the order goes in, it is arithmetic rather than opinion, and it does not appear anywhere on a payoff diagram.

Under the thin tailed model, the model that prices these very options, four structures over sixty expiries carry a 19.43 percent chance of halving the account and a 1.79 percent chance of losing all of it. Under the wider tailed model those become 25.78 percent and 6.42 percent. Two structures instead of four brings the halving probability to 6.42 percent; six structures takes it to 39.18 percent and the total loss probability to 16.59 percent. Position size is doing almost all of the work, which is what the arithmetic of an unbounded loss predicts.

The distribution free version of the same statement is cleaner, because it does not depend on any model at all. At four structures, the loss reaches half the account when the index settles at 30,217, a move of 20.87 percent, which is 4.85 standard deviations. It reaches the whole account at 34,384, a move of 37.54 percent. Those two index levels are facts about the position. Whether they are one in a hundred thousand events or one in four hundred events is the only open question, and the two models on this page answer it 309 times apart: 0.00084 percent against 0.258 percent per expiry. The payoff diagram is identical under both.

There is an obvious objection, and it deserves a computed answer rather than a hedge. If the market genuinely expected fatter tails, it would pay more for the calls being sold, and the structure would collect a larger credit. That is correct. Repricing both options consistently under the wider tailed model raises the credit from 50.60 points to 53.20 and moves the breakeven up by 2.60 points. Rerunning the entire ruin computation at the higher credit moves the sixty expiry halving probability from 25.78 percent to 25.01 percent, and the total loss probability from 6.42 percent to 6.15 percent. Being paid fairly for the tail is worth 0.77 percentage points of ruin over five years, about three quarters of a point. The reason is structural: the payment arrives in sixty small instalments and the loss arrives once.

How ruin arrives matters as much as whether it does. On the ruined paths, a single expiry accounted for more than half the entire fall from the account's high point in 93.7 percent of cases. The median gap between the equity peak and the ruin crossing was seventeen expiries, which sounds like ample warning until you notice that almost all of the damage happened in one of them. The account does not slide toward ruin. It sits at or near its high, and then one settlement moves it.

None of this is a claim about how often large index moves occur, which is not knowable from a simulation. It is a claim about sensitivity. A structure whose survival depends on an assumption about 4.85 standard deviation events is a structure whose survival is being decided by the least reliable part of any model. SEBI's September 2024 study of the derivatives segment found that 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. Structures whose risk is invisible in their own win rate are one of the mechanisms by which that happens.

Margin does not follow the payoff diagram

Everything above concerns what the position is worth. Margin concerns what the account must hold against it while it is open, and the two behave differently enough that the difference is worth its own section. No rupee margin figure appears here, deliberately: margin on Indian derivatives is computed daily from parameters the clearing corporation revises, and any number printed on a web page is stale the day after it is published. The mechanism, however, is stable and is what matters.

A defined risk spread attracts a margin benefit. The clearing system recognises that the long leg limits what the short leg can cost, so the requirement is closer to the actual maximum loss than to the notional of the short. That is why a vertical spread can be carried on a fraction of what the bare short would need. A ratio spread receives that benefit only on the covered portion. The second short call is not part of any recognised pair, so it is margined the way any uncovered short option is margined, which is against a scenario array of adverse moves in price and volatility, plus an exposure component on the notional. The result is a requirement that bears no fixed relationship to the credit received.

The consequence that catches people is the direction in which that requirement moves. Margin on an uncovered short is recomputed against a fresh scenario array each day, so as the index rises toward and through the short strike, three things happen at once and all of them are unfavourable. The scenario losses grow, so the requirement rises. The mark to market on the position is negative, so cash is debited. And if the move was violent, the volatility inputs widen too, which raises the requirement again independently of the price move. The account is asked for more capital at the exact moment it has less.

This is the same structural hazard that leverage creates anywhere, arriving through a different door. The difference is that a leveraged long position has a visible quantity of borrowing attached to it, whereas the uncovered leg of a ratio spread produces its leverage silently, as a by-product of the ratio. The number that makes it visible is the uncovered notional, Rs 15,30,000 per structure on the illustrative position here, and computing it takes one multiplication.

Two further mechanics belong in this section because both convert a manageable position into an unmanageable one without any decision being taken. The first is that a margin call met by closing part of the position can make the position worse rather than better, if the part closed is the long leg. Selling the long to raise cash leaves two uncovered short calls where there was one, doubling the slope of the loss arm at the worst possible moment. The second is that the covered half and the uncovered half expire together, which sounds obvious but means that any structure held into the final session carries its full uncovered exposure into a session where the settlement value is set by an average rather than by a price you can trade against.

The same view with a floor under it

The comparison that settles the argument is not ratio spread against nothing. It is ratio spread against the same directional view expressed with a floor under it. The view here is mildly bullish with an expectation of the index sitting near 25,500 at expiry, and the defined risk expression of that view uses the identical strikes: buy one call at 25,000, sell one at 25,500, and stop. That is an ordinary vertical, which our vertical spread page works through in detail and which this page uses only as the control arm.

Read the table honestly and the ratio spread wins most of the rows. It is paid to open rather than costing Rs 13,410. Its peak is twice as large. It finishes in profit nearly twice as often. On any scorecard built from frequency, it is the better structure, and a trader comparing them on those rows would reasonably choose it. The two rows where it loses are the last ones that matter and the only ones that are irreversible.

The vertical's own ruin figure deserves a mention rather than a footnote, because it is not zero. At four structures over sixty expiries it halved the account 2.09 percent of the time. A bounded loss repeated often enough still accumulates, and a page that presented defined risk as safe would be making the same mistake in the opposite direction. The difference is in how the two arrive. On the vertical's ruined paths, a single expiry accounted for more than half the fall only 55.9 percent of the time, and the median gap from equity peak to ruin was forty seven expiries against seventeen. The vertical grinds. The ratio spread detonates.

The sequence in the figure is one draw and should be read as one draw, not as a result. What it illustrates is the shape. The ratio spread won 48 of the 60 expiries, a rate of 80.0 percent against the 81.67 percent the distribution predicted, and reached a high of Rs 21,00,597 at expiry four. At expiry twenty-three the index settled at 29,423, a move of 17.7 percent, and that single settlement cost Rs 8,10,240. The account finished at Rs 11,49,484, down 42.5 percent, having been right four times out of five. The other fifty-nine expiries between them produced a net loss of Rs 40,277, which is worth noting because even the winning majority was not carrying the position.

The vertical spread on the identical sixty settlements won only 31 times and finished at Rs 23,53,700. That ending value is a draw from a distribution whose expectation is zero and should not be read as a property of the structure; the property to read is what happened at expiry twenty-three, where the vertical made Rs 65,583 from the same 17.7 percent move that cost the ratio spread Rs 8,10,240. Above its short strike the vertical's payoff is flat, so a large move and a small one produce the same result. Above the ratio spread's short strike there is no flat section at all. That is what the extra short leg bought, and that is what it cost. A settlement beyond that level is not a freak: under the wider tailed model, a sixty expiry sequence contains at least one settlement that large 21.6 percent of the time.

Three contracts, four bills, and what settles at the end

A ratio spread trades three contracts across two strikes, and pays a charge stack at every order. The arithmetic is worth doing because the conclusion runs against expectation, and publishing the inconvenient version is the point of computing anything.

Opening the position takes two orders. Buying one call at 497.60 turns over Rs 29,856 and costs Rs 35.95, made up of brokerage, the exchange transaction charge, stamp duty on the buy side and goods and services tax on the first two. Selling two calls at 274.10 turns over Rs 32,892 and costs Rs 85.55, the difference being the securities transaction tax of Rs 49.34, which falls on the sale of an option and not on its purchase. Entry therefore costs Rs 121.50. Closing at the peak costs Rs 103.71, because the long leg is then worth its full 500 points and pays transaction tax on the sale while the two shorts are bought back for a tick. The round trip is Rs 225.21.

Converted into index points that is 3.75, and as a share of the maximum profit it is 0.68 percent. The equivalent two leg vertical costs Rs 194.23 to round trip, so the third contract adds Rs 30.98. Here is the inconvenient result: the charge stack is not what makes this structure dangerous. It is a rounding error against a loss arm measured in lakhs, and any argument against the structure built on transaction costs is arguing about the wrong quantity by three orders of magnitude. About 41.9 percent of the bill is a flat per order fee that does not shrink with the trade, which does matter on cheap far out of the money variants, but not on this one.

Settlement is a different matter, and there are two routes with different bills. Index options in India are cash settled, so the uncovered short does not create a delivery obligation. What it creates is a cash obligation measured against the final settlement value, which is why holding to expiry rather than squaring off is a decision with consequences. Letting the position settle at the peak costs the exercise transaction tax on the long leg's intrinsic value, about Rs 45.00 here, against Rs 103.71 to close both strikes in the market, a difference of Rs 58.71. Against that stands the risk of carrying an uncovered short into the final session, which the expiry day page covers in detail.

Assignment on the uncovered leg behaves differently again depending on the underlying. For a cash settled index the obligation crystallises at expiry against the settlement value rather than arriving early, so what has to be managed is the settlement level, not the timing. For single stock derivatives the route is physical, and that mechanism has its own consequences for an uncovered short; it is set out on the page covering the risks of selling options and is not repeated here. The hazard common to both is the one already noted in the margin section: any sequence of events that removes the long leg while the shorts remain, whether through an early close, a partial fill or a leg by leg exit, converts the structure into a naked short position that nobody deliberately opened.

What a win rate cannot tell you

Every number on this page came out of one model of one structure, and none of it is a forecast. What generalises is the relationship between the numbers, and it is a relationship that holds for any structure with an uncovered leg.

A win rate is a statement about how often. It says nothing about how much, and in a zero expectation setting the two are locked together: raise the frequency of winning and you must raise the size of losing by the same factor, or the position would be worth more than it costs, which it is not. The 80.37 percent that makes this structure attractive and the Rs 8,50,277 worst expiry that makes it dangerous are not two separate facts about it. They are the same fact, stated twice.

The practical version of that is a habit rather than a rule. Any time a structure is described by how often it works, the missing half of the description is what happens when it does not, and the missing half can be computed before any capital is committed. Multiply the uncovered contracts by the strike by the lot to get the notional nothing is standing behind. Find the settlement level at which the loss reaches half the account, and convert it into a percentage move and a number of standard deviations. If that number is small enough to have happened in living memory, the structure's survival is a bet on the tail, whatever its win rate says.

The deeper point is about which quantities a model can be trusted on. The two distributions on this page agreed to within 1.29 percentage points on the win rate and disagreed by a factor of 309 on the probability that matters. Anything in the middle of a distribution is estimated from thousands of observations and is reasonably firm. Anything 4.85 standard deviations out is estimated from a handful and is mostly an assumption. Building a position whose survival depends on the second kind of number, while judging it by the first kind, is the specific error this structure invites. Working out which of your numbers are which, before rather than after, is the method we teach.

Every figure comes from one deterministic model, seeded at 20260815 so that it reproduces identically on each run. Option prices use the standard closed form model on an illustrative index at 25,000, thirty days to expiry, implied volatility of 15 percent held flat across strikes, a financing rate of 6.5 percent and no dividend adjustment. Quoted premiums are rounded to a five paise tick, which is the only reason the expectation comes out at Rs 18 below zero rather than exactly zero. The lot is illustrative and was chosen only so that contract value sits inside the current minimum contract value band, which the exchanges revise periodically; no lot size on this page should be read as the present one.

Two outcome distributions are used, and both price the at the money call at exactly the same premium. The thin tailed one is the lognormal the option prices imply. The wider tailed one is a two state volatility mixture in which one expiry in twenty is drawn from a state three times as wide, with the quiet state solved by bisection so the at the money price matches to within a millionth of an index point. That mixture has an overall standard deviation of 16.13 percent against 15.00 percent, an excess kurtosis of 4.65, and finite moments of every order, so its tail statistics are stable rather than dominated by a single draw. It is an assumption about shape, chosen to be moderately wider than the lognormal, and it is not a fitted model of any market. Distributional statistics use 5,00,000 expiries and the ruin computation uses 2,00,000 paths of sixty expiries each. Analytic checks on the breakeven, the peak, the win rate and the expectation are asserted in the script before any figure is written.

Charge rates are taken from the site's verified statutory research. Securities transaction tax on the sale of an option is 0.15 percent of premium and on an exercised option 0.15 percent of intrinsic value, both set by section 159 of the Finance Act 2026 with effect from 1 April 2026. Stamp duty on an option is 0.003 percent on the buy side, under Schedule I Article 56A(d) of the Indian Stamp Act 1899. The exchange transaction charge is modelled at Rs 3,250 per crore of premium turnover, the rate notified for index options by exchange notice 20240927-37 with effect from 1 October 2024. Goods and services tax at 18 percent is applied to brokerage and to the exchange charge, and not to the transaction tax or the stamp duty. Brokerage is an illustrative flat Rs 20 per order. One item could not be verified for the derivatives segment: the regulator's turnover fee is verified at Rs 10 per crore for the cash segment under regulation 41(1) of the 2026 broker regulations, and no derivatives figure is asserted here; at the cash rate it would add about Rs 0.06 across both entry orders, which changes no conclusion. No margin figure is quoted anywhere on this page, because clearing parameters are revised daily and a published number would be stale immediately.

All results are illustrative and simulated. They are not a track record, they are not a forecast, and they are not an indication of what any structure would produce in a live account. Nothing here is advice to enter, avoid, hold or adjust any position, and no part of this page should be read as a view on whether any structure is suitable for any person.

• The vertical spread, computed in full →
• Risk of ruin, and the three levers that set it →
• What leverage actually does to an account →
• The options framework, from first principles →