Guide · Execution mechanics · Updated 2026-09-19

One guarantees execution and not price, the other guarantees price and not execution

Most writing on this choice offers a preference. The difference is not a matter of taste and is not decidable in the abstract. It is decidable from the distribution of price jumps in the market you actually trade, and that distribution can be read off the exchange's own daily files.

A guarantee of execution, or a guarantee of price, and never both

Both order types carry a trigger, which holds an order dormant until the market prints at or through a stated price and then releases it into the book. Everything that separates the two happens after release.

A stop-loss market order releases a market order, which crosses the spread and consumes resting liquidity until it is filled. It cannot fail while there is anything on the other side, and it accepts whatever prices that consumption reaches. Execution guaranteed, price uncontrolled.

A stop-loss limit order releases a limit order at a second price you nominate. A limit order never trades worse than its price, so if the book holds nothing at or better, it rests until cancelled or until the session ends. Price guaranteed, execution not. A reader holding those two sentences can derive the rest of this page.

The failure modes are exact opposites, which is why neither is safer

A market variant fails by filling badly. You wanted out near a stated price and got out well below it. The loss is realised, larger than planned, and over.

A limit variant fails by not filling. You are still holding, further from that price than when the trigger fired, with no protection running. The loss is unrealised, open, and continuing.

These are opposite quantities, not two degrees of one thing. One converts an unknown future loss into a known present one at a bad rate; the other declines the conversion. Anyone calling one of them the safe choice has answered a question about your book that they did not ask.

The only thing that separates them is a price that skips

A price moving continuously through a level touches every price on the way, which is why a limit order below a falling market fills: the market arrives at it. A price that jumps does not touch what it jumps over, and a limit sitting in the skipped region is simply passed. Jumps come from three places here: the overnight break, the intraday jump on news that daily bars cannot see, and a price band. Take the first on the broad benchmark index.

The median absolute opening difference is 0.272 per cent and the ninety fifth percentile 1.102 per cent. An index is a portfolio, so single securities jump considerably more, but even at index level a gap beyond one per cent arrives about once every three weeks. A stop placed inside that distance will regularly be resolved by an opening print rather than by trading.

Now the individual security measurement. The daily bar answers the question exactly in one case: if a session opens at or below a sell stop, the trigger fires at the opening print, the entire session therefore lies after the trigger, and the session high is the post-trigger high. A stop-loss limit could have filled only if the high reached it. No assumption about the intraday path is needed.

Read the last two columns as a pair: they are the two failure modes side by side. At a 2 per cent stop the market variant filled 1.63 per cent below the trigger on average across gap-through sessions, while the cases a limit at the trigger did not fill were still open and 4.91 per cent below it by the close. Push the stop to 5 per cent and the market fill worsens to 2.99 per cent while the unprotected position worsens to 6.24 per cent. Both grow with distance, by a similar 1.36 and 1.33 points, and at every distance the position a limit left open is the worse of the two.

Note what does not happen. A wider stop is triggered far less often, 5.83 per cent of observations against 57.56 per cent for a one per cent stop, but conditional on triggering its chance of being gapped through falls far less, 5.85 per cent against 12.65 per cent, so roughly one trigger in 17 still arrives by a jump. Distance buys fewer triggers. It does not buy an orderly one.

Two prices, and setting them equal is the common way to break the order

A stop-loss limit carries two prices, and a large share of the orders entered in this market carry them at the same value. That is not a conservative setting. It is the setting that makes the order least likely to act in the only case where it differs from the alternative, because a gap-through then releases a limit priced above where the market now is, and it fills only if the price comes back.

Three readings, in order of how often they are missed. First, on the mean the limit order beats the market order at every band, because in most gap-through sessions it fills at its limit rather than at a worse opening print. A page that stopped at the mean would recommend it.

Second, the tail reverses that. With the limit at the trigger, the fifth percentile outcome is -5.89 per cent against the market order's -5.51 per cent, and 6.43 per cent of cases are worse than five per cent against 5.82 per cent. The order with the better average has the worse bad day. Widening the gap moves the tail back across that line: from 0.50 per cent under the trigger onward the limit order's fifth percentile is no longer worse than the market order's.

Third, every unfilled case in that table is marked at the same session's close as though the position had been closed there. It had not been.

A non-fill is not a near miss, it is an open position

When a stop-loss limit does not fill, nothing has gone wrong mechanically. The order is working as specified. What has happened is that protection has quietly become a preference, in the one session where the position moved most. Follow those cases forward.

By the next session's close 85.3 per cent were still below the trigger and 44.9 per cent more than five per cent below, with a fifth percentile of -14.55 per cent. Five sessions on the fifth percentile is -21.59 per cent and the worst case -92.7 per cent. The market variant had exited at the open, on average -1.63 per cent from the trigger, and was exposed to none of it.

This is the selection effect the averages hide. A limit fails to fill precisely in the sessions where the price kept going, because those are the sessions in which it never came back to touch the limit. It is not a random subset of gaps. It is the subset in which the decision mattered.

The most traded names are where the limit fails most

Folk practice says use a limit where the book is thin and a market order where it is deep. The data says the second half is backwards. Split the same universe into thirds by that session's turnover.

In the most traded third a limit at the trigger failed on 25.30 per cent of gap-through sessions against 12.85 per cent in the least traded third, a little over half as often. The mechanism is the recovery column: after gapping through, a thinly traded security rose a median 3.81 per cent off its opening print against 2.87 per cent for a heavily traded one. An opening print struck on thin interest is a noisy estimate the session then trades away from. One struck on heavy interest is close to a considered price, and the security stays near it.

The tail columns complete the inversion. In the most traded third the limit's fifth percentile, -7.96 per cent, is worse than the market order's -6.96 per cent. In the least traded third it is better, -3.61 against -4.41 per cent.

One qualification, which strengthens this rather than weakening it. A market order does not fill at the opening print; it fills by consuming the book from that print. The figures above use the print, so they bound how good a market fill could have been, and the bound is loosest exactly where the book is thin. The real market outcome in the least traded third is therefore worse than shown, while in the most traded third the printed figure is close to attainable. Both corrections push the same way as the table.

Price bands, where both order types fail together

This is the condition generic treatments omit, and here it is not exotic. Securities carry price bands, and securities in the derivatives segment carry an operating range that can be flexed intraday against objective criteria rather than a fixed daily limit. Within a band nothing trades outside the band, by construction.

Take both order types into that state. A limit priced outside the applicable band cannot execute, and depending on the venue and the moment it is either rejected on entry or cancelled when the band moves. A market order sent when a security sits at its lower band with no bid has nothing to consume, and depending on segment and product it may be rejected, converted, or simply sit. The situation that most resembles the emergency a stop was bought for is the one in which both variants can be inert.

One India specific detail changes what survives a band move. When a dynamic price band is flexed, pending orders whose limit prices now sit outside the new band are cancelled by the exchange, and untriggered stop-loss orders are carved out of that cancellation. So an untriggered stop survives a flex, while a stop-loss limit that has already triggered is an ordinary limit order exposed to the same cancellation as any other. In a banded move, triggering can be the moment your instruction becomes cancellable.

792 observations had a single price for an entire session after a fall past four and a half per cent. Rare as a share, not rare as an event: several hundred occasions in under five years, and whoever held those positions met the limit of an order type on the day it mattered. How Indian price bands are set and revised is the companion to this section.

The last stretch of the cash session stopped being continuous

Here is the change that dates most of what is currently published on this topic. In January 2026 the market regulator introduced a closing auction session in the equity cash segment, effective 3 August 2026, initially for cash market securities on which derivatives contracts are available. For those securities the closing price is no longer a volume weighted average of the last half hour of continuous trading. It is an equilibrium price struck in a call auction.

Continuous trading in them now ends at a quarter past three. What follows is an auction with a published phase structure: a reference price window, a period in which the reference price is disseminated and no auction orders are accepted, an order entry phase taking market and limit orders, then a limit only phase whose entry freezes at a randomised moment before matching. A revised pre-open framework followed from 7 September 2026.

The detail that matters is the order type list. Only market and limit orders are permitted in the auction. Stop-loss orders are not, and unlike unexecuted limit orders, which carry in with their time priority, a stop-loss order does not carry across. For a security with derivatives contracts there is now a stretch at the end of the cash session in which a resting stop has no mechanism to act against, and in which the official closing price of the day is determined without it.

Three consequences follow. Any instruction that depended on a stop firing in the closing minutes of those securities no longer has a session to fire in. Any automated square off timed against the old continuous close is timed against a boundary that moved. And any article, course or checklist describing the last half hour of the cash session as continuous trading for those securities was written before August 2026. The mechanics of a call auction and the structure of the Indian trading day sit either side of this one.

The condition that should decide, read off the measurement

Label every session by the volatility of the broad index over the previous 21 sessions, with thresholds re-estimated each session from earlier readings only, so nothing in the label uses information from the session it labels. Then run the same stop through it.

The gap-through rate roughly doubles from the calm third to the stressed third, 5.66 per cent against 11.83 per cent, and the outright failure rate rises from one trigger in 93 to one in 53. That is the volatility condition, measured.

There is a subtlety in the middle column that deserves attention rather than suppression. Conditional on a gap-through, the non-fill rate is lower in the stressed third, 15.94 per cent against 19.07 per cent, because a stressed session has a wider range and is more likely to trade back up to the trigger at some point. The two effects oppose each other, the gap rate dominates, and the joint failure rate still rises, but by considerably less than the gap rate alone suggests. Quoting only the gap rate overstates the case.

Put the two readings together and the conditions separate cleanly. The limit variant's specific weakness, a refusal to fill, concentrates where opening prints are informative and hold: heavily traded securities, and by implication the ones with derivatives and tight books. The market variant's specific weakness, an uncontrolled fill, concentrates where the book is thin and where volatility is high enough that the first available price is far from the trigger. What the measurement does not do is choose, and a page producing a setting from this table would be inventing the missing input, which is the size of the position relative to the account. Sizing decides what a bad fill costs.

A stop is an instruction to a matching engine, not a floor under a position

Everything above assumes the instruction reaches a working order book, is accepted, and meets a counterparty. Each can fail independently of the order type. The trigger evaluates against the exchange price feed, so a stop entered as an exchange order type is bound by exchange rules on which order types a session accepts, which the closing auction change shows is not fixed. Wrappers layered on top, including the bracket and cover products many Indian brokers offer and any alert-then-place arrangement, may hold the trigger at the member, where connectivity, risk policy and square-off timings apply instead. These are different instruments wearing the same name.

Then the part daily data cannot measure. Everything quantified here concerns overnight jumps, because the daily bar answers that exactly. Intraday jumps exist too, and a fast move that skips ticks inside a session produces the same non-fill with no gap in the record, so every non-fill figure here is a floor on the true rate rather than an estimate of it. Meanwhile the excursion measurement shows how far past a trigger a session runs even when it triggers in ordinary trading: a median 1.16 per cent below the trigger, 5.25 per cent at the fifth percentile and 8.59 per cent at the first. A stop does not stop a price.

A stop is a conditional instruction to a matching engine, and what it can achieve is bounded by what that engine is doing when the condition is met. The order type chooses which of two failure modes you are exposed to. It removes neither. Modelling that honestly in a backtest is where most of the damage is done, because a backtest filling every stop at its trigger price has assumed away this entire page.

Reading an order type as a guarantee, then finding the sentence that names what it guarantees and what it does not, is the judgement that separates a working method from a collection of settings.