Guide · Market structure and risk · Updated 2026-09-23

A market-wide halt stops the forced sale for an hour, not the margin call that forces it

The usual account of March 2020 says the circuit breaker stopped the crash. The daily record neither confirms nor refutes that, because a halt was never built to set a price. It is built to interrupt one step of a mechanism for about an hour, and the mechanism kept running on both sides of that hour. This page states the mechanism, checks what the rules did on each halt day, and measures the month from the exchange's own files: daily index closes and ranges for 3,385 sessions from 2013, India VIX, eleven sector indices, and participant-wise open interest in index futures for every session of the period.

The loop runs through the clearing system, not through sentiment

A leveraged long position meets a falling market through five steps, and none of them requires anyone to panic. The fall is marked to market at the close, so the loss comes out of the account's equity that evening. The same fall enters the volatility estimate from which margin is sized: the exchange's published VaR method keeps 94 per cent of the previous day's estimate and adds 6 per cent of the latest squared return, and the price scan range for index derivatives is a multiple of the volatility computed for VaR. So the requirement per contract rises, and so does the haircut on pledged shares, which the exchange's margin reporting rules set at no less than the security's VaR margin rate at the start of the day. Equity falls below the requirement. Unfunded, the position is closed by the holder or the broker, into the same market, and the price falls again.

Two properties make the loop amplify a fall rather than absorb it. It is procyclical: the requirement rises after a large move, so it asks for the most collateral when collateral is worth least. And it lags: an estimate that keeps 94 per cent of yesterday has a half-life of 11.2 sessions, so it peaks after the price and stays high into the rebound. Every step is prudent on its own. A clearing corporation that held margins flat in a volatile market would be carrying its members' risk; a broker that left an unfunded position open would be carrying the client's loss. The instability is the sum of prudent steps taken at once, in one direction, against one order book.

A halt stops matching and cancels every resting order

The market-wide breaker comes from SEBI's circular of 28 June 2001 and takes its present shape from the circular of 3 September 2013. It applies at moves of 10, 15 and 20 per cent either way in the Sensex or the Nifty 50, whichever is breached first, with limits computed each day from the previous close, and it halts all equity and equity derivative markets nationwide at once. A 10 per cent breach before 1 pm halts trading for 45 minutes, a 15 per cent breach before 1 pm for 1 hour 45 minutes, and a 20 per cent breach at any time closes the market for the day; later breaches bring shorter halts or none. The full matrix is set out in the guide to India's three circuit systems and the introduction to circuit limits. Every halt short of the rest of the day ends with a 15 minute pre-open call auction.

Two rules decide what a halt does to a position. The first is cancellation. The consolidated circular of the Metropolitan Stock Exchange of India of April 2024, restating the SEBI circulars of 2001 and 2013, says that all outstanding orders at the time of a halt caused by the market-wide breaker shall be cancelled, and SEBI's circular CIR/MRD/DP/02/2015 of 12 January 2015 sets the mechanics: the exchange recomputes the index after every trade in its constituents, stops matching the moment a limit is breached, and purges the unmatched orders. A stop order resting on the exchange at 9:19 am on 13 March did not exist at 9:21. The second is the reopening. The first price after a halt comes from a call auction, not from the last trade, so a position is marked at a price nobody could trade at when the halt began.

The breaker is also symmetric, which much writing on it overlooks. It fired on a rise on 18 May 2009, twice in the same session, and closed the market for the day. That is why the line repeated on 13 March 2020, that trading had halted for the first time in twelve years, was inaccurate: it was the first halt on a fall since 22 January 2008. One consequence of a halt is recent. Since 3 August 2026 stocks with derivatives close in a closing auction, and on a day the market shuts early on an index breaker no auction is held and those stocks close by the older method (exchange circular NSE/CMTR/74466, set out in the guide to the closing auction).

Two halts, six sessions apart, and opposite days

The two days differ in exactly what the rule cannot see. On 13 March the broad index opened 5.03 per cent down, crossed the trigger within five minutes of the open, and printed a low 10.79 per cent below the previous close. That low sits below the 8,625 reported at the moment of the halt, so some of it was printed in the reopening auction or after it, which a daily file cannot separate. The index then rallied to close 3.81 per cent up. Its range that day, 16.73 per cent of the previous close, is the widest in the 3,385 sessions of the file record and nearly twice the next widest, 8.77 per cent on 19 March 2020.

On 23 March most of the damage was done before the first trade. The index opened 9.14 per cent down, 69 per cent of the day's eventual fall and less than one percentage point short of the trigger. The breaker fired at 9:58 am, the call auction reopened trading at 10:58 am, and the selling resumed: the close of 7,610.25 was 12.98 per cent down, 3.31 per cent below the trigger level and 0.35 per cent above the day's low. It was the lowest close of the episode. The day before had been Sunday 22 March, the day of a national curfew, and the 21 day national lockdown was announced on the evening of 24 March.

That opening gap is the general case. From the close of 20 February to the close of 23 March the index fell 37.01 per cent over 20 sessions, and 67.0 per cent of that fall, measured in logs, arrived between one session's close and the next session's open, when no breaker operates. A halt is a rule about continuous trading. A market that reprices overnight reaches it only after the gap.

The record of the month

March 2020 held 11 sessions that moved 3 per cent or more; the whole of 2019 held 2. India VIX, the exchange's index of volatility implied by index option prices, closed at 13.70 on 20 February and at 83.61 on 24 March, 6.1 times higher and the highest close in the files since they begin on 14 May 2014, with an intraday high of 86.64 the same day. The average March session covered a range of 5.47 per cent of the previous close, five times the 2019 average of 1.09.

The sector columns put a number on the phrase that correlations go to one in a crisis. Cross-sector dispersion rose from 0.97 points in 2019 to 1.87 in March, but the broad index's average daily move rose 6 times over, so relative to the size of the move the sectors had never moved more alike, and their mean pairwise correlation rose from 0.44 to 0.85. In April dispersion held at 1.86 while correlation fell to 0.62. The sectors fell together; they did not rise together.

The margin followed the volatility, and the estimate lags by design

Run the same estimator over the broad index's daily log returns and the margin's core input is on the page. On 20 February it stood at 0.84 per cent a day. It reached 1.29 on 28 February, 2.70 on 12 March and 4.87 on 23 March, and peaked at 4.88 on 25 March, 5.8 times its February level and two sessions after the price trough. It stayed within 10 per cent of that peak until 13 April, while the index rose 18.2 per cent off its low, and at the end of June it was still 1.93 times its February level.

The lag is deliberate: an estimator that forgot a shock the day after would under-margin the next one. Its consequence for anyone levered is mechanical. The exposure a given equity can carry is equity divided by the margin rate, and in a fall both move against the holder. Take an illustrative schedule of the published form, three times the estimate scaled to a two day horizon with a floor of 5 per cent of notional; the parameters are chosen for illustration and are not those in force in 2020. The requirement first rises above the floor on 28 February, then climbs to 20.7 per cent on 25 March, 4.1 times its February level, so the exposure one rupee of equity could carry fell from 20 times to 4.8. The fall of 23 March alone consumed 85 per cent of the requirement set the evening before.

Collateral shrinks on the same schedule. Pledged shares count at their value after a haircut of at least their VaR margin rate, so pledged holdings lose collateral value twice in a fall, once through the price and again through the larger haircut. Both sides of the account move against the holder on the same input.

What a leveraged account faced on a halted market

In the hour after 9:20 am on 13 March a holder of leveraged index futures could not sell, could not hedge on an Indian exchange, because the derivatives market was halted in the same instant, and had lost every order resting on the exchange. What remained possible was to add money. The requirement did not pause, and the position would reopen at an auction price.

To see where the loop bites, run an illustrative account on the real closing path: long index futures at 2, 3, 4 and 5 times equity from the close of 20 February, marked to every close, with any shortfall met by selling exactly enough at the close to bring the requirement back to equity. There are no top-ups, no re-leveraging, no slippage and no intraday path, and every one of those simplifications flatters the account.

The 5 times account was wiped out at the close of 12 March, before either halt: the breaker had nothing left to protect. The 2 times account was barely touched. Between them the margin decides the outcome, and not in the direction intuition suggests. With the requirement following volatility, the 3 times account was forced to sell on 16, 17, 18, 19 and 23 March and came through the trough with 5.3 per cent of its starting equity. With the requirement frozen at its February level it sold later, carried more exposure into the worst session, and was wiped out on 23 March, the second halt day. The volatility-scaled margin did what it exists to do, which is to force the cut early enough to keep the account solvent. The same cut, made by every leveraged account at once, is the selling that deepened the fall.

Survival is not recovery. At the trough the forced sales had left the 3 times account with 13.6 per cent of the exposure it would have carried unforced. The index then rose 35.4 per cent by the end of June, and the account finished June at 14.4 per cent of its starting equity, against 85.3 per cent for an unlevered holder of the same index. The unlevered holder waited out a drawdown; the levered one realised a loss at the bottom and was absent from the rebound. That asymmetry, rather than the depth of the fall, is what scaling exposure to volatility tries to get ahead of. The measurement has also tightened since: under SEBI's circular SEBI/HO/MRD2/DCAP/CIR/P/2020/127 of 20 July 2020 the requirement has been checked at intraday snapshots from 1 December 2020, in full from 1 September 2021, so a halt day's intraday move now reaches a shortfall directly, as the peak margin guide and the shortfall penalty guide work through.

Who sold: the positions data

The exchange's participant-wise open interest files record, for every session, the index futures contracts held long and short by clients (everyone who is not an institution or a broker's own book), domestic institutions, foreign portfolio investors (the file's FII column) and brokers' own books. Counts are contracts across every expiry, not rupees, and the March contracts expired on 26 March.

Clients went into the slide long and kept adding. Their long index futures rose from 179,456 contracts on 20 February to 235,173 on 6 March, and then came out: 82,506 on 24 March, 64.9 per cent lower, before the expiry. The three largest single-session reductions fell on 24 March (36,320 contracts), 23 March (29,400) and 13 March (22,387): the two halt days and the session after the second, which together carried 57.7 per cent of the reduction. Foreign portfolio investors held the other side. Their short futures peaked at 208,785 on 6 March, fell 38.5 per cent between 20 and 24 March, and on 24 March their net position turned long for the first time since 21 January. Client net positions, long on every session from 20 February through 23 March, turned short the same day.

The pattern is what the loop predicts: leveraged longs closed into a falling market and bought back by shorts positioned for the fall. The file cannot prove it. It does not say whether a contract was closed by a margin call or by choice, and a category hides offsetting accounts inside it. What it does show is that the selling that ended on 23 and 24 March came disproportionately from the category that had levered up on the way down. The guide to flow figures explains why the headline foreign flow number, which is a cash market figure, would have missed this derivatives side entirely.

What SEBI changed that week, and what it did not

SEBI's press release 18/2020 of 20 March 2020 observed that the settlement cycles of the exchanges and clearing corporations had not been disrupted despite the market's movement, and introduced five measures from the start of trading on 23 March, for one month subject to review.

None of the five prohibited short sales in the cash market, whatever the headlines of that weekend said. The constraint fell on index derivative positions and on the leverage available in volatile stocks. SEBI extended the package through 2020, withdrew the derivative stock and cash margin measures with effect from the close of business on 26 November 2020, and kept the index derivatives limits and the cooling-off before flexing in force until further directions. The ban-period rule that governed the halved limits has itself changed since, as the guide to the ban period sets out.

Monetary support followed on 27 March. The Reserve Bank cut the repo rate by 75 basis points to 4.40 per cent and the reverse repo rate by 90 basis points to 4.00 per cent, cut the cash reserve ratio by 100 basis points to 3 per cent, and put the liquidity released by its term repo, reserve ratio and standing facility measures at 3.74 lakh crore rupees (Statement on Developmental and Regulatory Policies, 27 March 2020).

Whether the halt worked is a question a daily file can only partly answer

Rank every session in the file since 2013 by how far the broad index fell below the previous close at its low, and take the 12 deepest. Both halt days are there, and on the one statistic that describes what happened after the worst moment they sit at opposite ends.

13 March recovered 135 per cent of its fall by the close, the most of the 12; 23 March recovered 2.3 per cent, the least. The other 10 recovered a median of 16.8 per cent, from 4.1 per cent on 16 March 2020 to 79.4 per cent on 9 November 2016. The sessions after the halts were not calmer either: 16 March fell 7.61 per cent, and 24 March, which rose 2.51 per cent, carried the highest India VIX close in the files. The 10 other extremes were followed by a median absolute move of 2.06 per cent.

What a daily record can say is limited but firm. The halts did not prevent the lowest close of the crash, which came on a halt day, and they were not followed by calm sessions. On everything a daily file carries, the halt days are indistinguishable from other extreme days. What it cannot say is whether either halt changed the outcome. There is no counterfactual day without the halt. The effect a halt is suspected of adding, a magnet in which trading accelerates as the index nears the trigger because participants rush to act before they are locked in, is a statement about the minutes before 9:20 and 9:58, invisible in daily bars. The week of 22 to 27 March also carried a national curfew, the SEBI measures, the lockdown announcement and the Reserve Bank's package, and any comparison absorbs all of them. Two events are not a sample. The Indian data that could answer the question is the exchange's intraday order and trade record around the two triggers.

The narrower reading the daily files allow is that the loop ended when its fuel ran out. Client long futures, the position the loop feeds on, had fallen by close to two thirds by 24 March. The index's lowest close came on the 23rd, its first rise on the 24th and a 6.62 per cent rise on the 25th. That sequence is consistent with forced selling exhausting itself; it is equally consistent with a policy response arriving in the same week, and the data cannot divide the credit.

The fall was shared; the recovery was selective

The pharmaceuticals index regained its 20 February level after 10 sessions, on 8 April 2020; energy and consumer staples after 69; information technology after 77; automobiles, which bottomed on 3 April, after 88. The broad index took 156 sessions, to 5 November 2020. Banks took 192 and financial services 179. Public sector banks and realty did not bottom until 19 May, and took 182 and 155 sessions from there. Media took 294, to 10 June 2021, and on 18 September 2026 stood 17.0 per cent below its 20 February 2020 close, the only one of the fourteen below its pre-crash level today.

In March the sectors fell together; from April they separated, and the separation lasted more than a year. None of that ordering was knowable on 23 March. A list that looks obvious now is the record of what the rebound did, not a forecast anyone could have traded.

The lesson, stated as mechanism rather than hindsight

Five things follow from the mechanism, and none of them needs knowledge of how March 2020 ended.

Leverage decides whether the loop reaches you, and it is decided before the fall. In the illustration nothing forced a 2 times account, and a 5 times account was closed before the first halt.

Fund the requirement at the peak of the estimate, not today's. The requirement is a lagging function of realised volatility. It rose 4.1 times in the illustration and stayed near its peak for three weeks after the price turned.

A halt suspends your ability to act, not your exposure. Resting orders are cancelled, the derivatives market is shut, and the position reopens at an auction price. The one action left inside the halt is funding.

Forced selling turns a drawdown into a realised loss at the worst price. It also removes the position from the recovery. Coming through the trough with a sliver of exposure is a different outcome from holding through it.

Most of a crash can arrive overnight, where no breaker operates. Of this one, 67 per cent did. A stress test that models only intraday moves misses the step that did the damage; the stress testing guide reprices a book through this window and others, and the tail risk guide measures why a standard deviation fitted before March 2020 could not see it coming.