Guide · Market history and method · Updated 2026-09-23

The index fell 5 per cent in demonetisation's first week, while the sectors inside it ranged from minus 20 to plus 10

Most retellings of demonetisation's market day quote the intraday low and call it a crash. In the exchange's own session file the opening print, the close and the spread between sectors say three different things, and every number below can be redone from the method in the closing note.

The crash was in the opening print, and the close was an ordinary bad day

Measured from its 8 November close, the broad index opened 5.57 per cent down on 9 November. Across 3,373 single session returns in the files since 2013, only one opening print was lower, on 23 March 2020. The day's range from low to high was 5.55 per cent of the prior close, wider than all but 0.36 per cent of sessions. Then the index closed 1.31 per cent down, a result that 6.3 per cent of sessions have beaten on the downside.

India VIX told the same split story. It opened at 16.77, traded as high as 22.79, a jump of 35.9 per cent that only 0.30 per cent of sessions have exceeded, and closed at 16.62, below the previous day's close. Any summary built from closing prices says little happened. Any summary built from the opening print or from the cross section says something extreme did. Both are true, and the gap between them is the first lesson of the event.

The figure shows where the extremity went. At the open, ten of the eleven sector indices sat inside a band 2.12 points wide, from -7.36% to -5.24%. An opening print is the first price after a night of news, set before the market has traded through it; for the broad index's constituents it comes from the pre-open call auction, which has covered them since October 2010 and clears each stock against the same overnight information at once. By the close those ten spanned -3.25% to +2.10% and four of them had closed higher than the day before, while realty opened 11.48 per cent down and closed 11.60 per cent down. The cross sectional standard deviation of the eleven moves doubled during the session, from 1.72 points to 3.52. The market did not fall and recover. It fell together, and then it sorted.

One withdrawal, two channels pointing in opposite directions

The legal instrument was Gazette notification S.O. 3407(E) of 8 November 2016, under which the ₹500 and ₹1,000 notes of the existing series ceased to be legal tender with effect from 9 November. The Reserve Bank's circular of the same date, DCM (Plg) No.1226/10.27.00/2016-17, set the operating terms: 9 November was a non business working day for banks; ATMs and cash machines were shut on 9 and 10 November; old notes up to ₹4,000 could be exchanged over the counter; cash withdrawal from a bank account was limited to ₹10,000 a day and ₹20,000 a week until 24 November; ATM withdrawals were limited to ₹2,000 a day per card until 18 November; and old notes could be paid into an account without limit where know-your-customer compliance was complete, until 30 December. What was withdrawn was large: notes valued at ₹15.4 trillion, which the Reserve Bank's November 2017 Bulletin put at 86.9 per cent of the value of all notes in circulation.

Those terms opened two channels, and the equity market felt both. The first is a cash channel. Where a purchase can only settle in notes, rationing the replacement notes removes the medium of exchange, so the sale does not happen at any price; that is a mechanical stop rather than a change of mood. The Reserve Bank's March 2017 assessment, Macroeconomic Impact of Demonetisation, records two-wheeler sales 22.0 per cent lower in December 2016 than a year earlier. The second is a deposit channel. Unlimited deposits meant the withdrawn notes did not vanish; they became bank deposits. Between 28 October 2016 and 6 January 2017 currency in circulation fell by about ₹8,800 billion and aggregate deposits rose by about ₹6,720 billion, according to the same assessment. The surplus had to be placed at once, in government bonds, where the ten year yield fell from 6.80 per cent on 8 November to 6.18 per cent on 24 November, and with the Reserve Bank.

The Reserve Bank's absorption measures the flood. It imposed an incremental cash reserve ratio of 100 per cent on the increase in banks' net demand and time liabilities between 16 September and 11 November 2016, effective from the fortnight beginning 26 November, and withdrew it from the fortnight beginning 10 December once the ceiling for Market Stabilisation Scheme securities had been raised to ₹6,000 billion (press release 2016-2017/1443, 7 December 2016). At the peak, on 4 January 2017, it was absorbing ₹7,956 billion through reverse repos and cash management bills. The overnight rate barely moved: the exchange's one day rate index implies an average overnight rate of 6.23 per cent over the twenty sessions before the announcement and 6.07 per cent over the twenty after, inside a corridor running from the reverse repo rate of 5.75 per cent to the marginal standing facility rate of 6.75 per cent around a repo rate of 6.25 per cent. A flood does not show in an overnight rate the central bank will absorb at a floor. It shows in the quantity absorbed and in rates that have no floor, which is why bonds rather than the money market carried the signal.

The final count settles what the notes turned into. The Reserve Bank's Annual Report for 2017-18 put the notes returned, after verification, at ₹15,310.73 billion, against the ₹15.4 trillion withdrawn: more than 99 per cent of the value came back. The money changed form, from notes to deposits, which is why the financial footprint was a liquidity surplus and a bond rally. The cash has since been more than replaced. Banknotes in circulation were valued at ₹41,23,995 crore at end March 2026, according to the Annual Report for 2025-26, about 2.3 times the roughly ₹17.7 lakh crore of notes in circulation on 8 November 2016, and the ₹500 note alone carried 85.5 per cent of that value, against 86.9 per cent for the two withdrawn denominations together. The ₹2,000 note, introduced in November 2016 to meet the currency requirement quickly after the withdrawal, was itself withdrawn from circulation under a Reserve Bank announcement of 19 May 2023; it remains legal tender, and 98.45 per cent of it had come back by 31 March 2026. A Constitution Bench of the Supreme Court upheld the 2016 notification by a four to one majority on 2 January 2023. A page that still describes demonetisation as a durable reduction in cash is describing a position the Reserve Bank's own figures have reversed.

Two shocks inside one session

The announcement came on the evening of 8 November, after the Indian close, and the 8 November session file shows an ordinary day, the broad index up 0.55 per cent. The United States presidential election was held the same day. Polls there began closing at 6 pm Eastern time, 4.30 am on 9 November in India, the count ran through the Indian morning, and the result was called at about 2.30 am Eastern, about 1 pm in India, in the middle of the Indian session. The pre-open auction from 9 am therefore priced the domestic news in full and the foreign count in progress, and the 3.30 pm close priced both.

So the first day's index move is a joint response and cannot be split. The multi week path is worse, because the global repricing that followed the United States result kept going: on the Treasury's own daily par yield curve the ten year yield was 1.88 per cent on 8 November, 2.07 per cent on 9 November and 2.36 per cent on 23 November, and the Federal Reserve raised its target range on 14 December 2016. In between, the Monetary Policy Committee held the repo rate at 6.25 per cent on 7 December. Each is a further event inside any window long enough to show a recovery.

There is one market where the two shocks predict opposite signs, and that is enough to separate them there. A global shock that lifts world yields pushes Indian yields up; a domestic deposit surge pushes them down. United States ten year yields rose 48 basis points from 8 to 23 November, and Indian ten year yields fell 62 basis points from 8 to 24 November. The exchange's ten year government bond index gained 0.93 per cent on the first session and 2.65 over five, a five session gain that only 0.07 per cent of windows since the index's first file in 2015 exceeded, and the largest five session gain anywhere in its record, 3.15 per cent, began at the close of 11 November 2016, inside the same episode. In Indian government bonds the domestic channel won outright.

The currency and the equity index cannot be separated this way. The rupee value implied by the ratio of the dollar denominated broad index to the rupee one fell 2.93 per cent by 28 November 2016, which a stronger dollar and a cut in Indian yields would both produce, so neither cause is assigned.

The sector cross section can be read only once an assumption is added: an exposure map stating which sectors each shock should touch. Realty's abnormal move on the first day, -9.69% after allowing for its usual relation to the market, has an obvious domestic cash channel and no United States channel of that size. The two export heavy sectors, information technology and pharmaceuticals, moved in opposite directions the same day, -2.20% and +2.98% abnormal, which a domestic cash shock does not explain and a change in United States policy expectations could. Those readings come from the map, not the data. The deeper problem is structural: the natural control group for a domestic cash shock is the sectors with no domestic cash exposure, and the export sectors are exactly the ones a foreign shock hits hardest. This natural experiment has no clean control group.

What each sector did, measured from its own close

Five sessions in, the ranking shows the cash channel and the deposit channel side by side. Realty fell 20.23 per cent, automobiles 11.52, consumer staples 8.11 and the thematic consumption index 8.89, against 5.06 for the broad index, while public sector banks rose 9.75. The Reserve Bank's March 2017 assessment, measured on the other exchange's indices over what it describes as 9 November to 30 December, found the same order: realty down 14.4 per cent, consumer durables 9.9, automobiles 9.0 and consumer staples 5.3, against 3.5 for that exchange's benchmark. Measured from the 8 November close to 30 December, this exchange's indices give realty -16.57%, automobiles -9.31%, consumer staples -5.24% and the broad index -4.19%. Two index families, one ordering.

The banking split is the clearest trace of the deposit channel in equity prices, and it shows the arithmetic by which a bond rally reaches a bank share. On the eve, the file's own price to book column put the public sector bank index at 0.96 times book and the private bank index at 2.86. A gain on government bond holdings adds to book value. For holdings of similar size relative to book, the same gain is about three times larger relative to market value at 0.96 times book than at 2.86. The public sector bank index closed the first session up 2.10 per cent and the private bank index down 0.34. The arithmetic is consistent with that split. It does not prove it, and a price file cannot.

Unusual against what: every ordinary window as the comparison

An extreme number needs a distribution to be extreme against. The comparison here is every window of the same length in the exchange's files from 2013-01-01 to 2026-09-18, measured the same way: the standard deviation of the eleven sector returns from the window's first close to its last. Nothing is sampled.

The claim that dispersion is the unusual part of this event holds, and then it fails. Over one session the spread was wider than all but 0.27 per cent of sessions, and over five sessions all but 0.42 per cent, 3.5 times the median five session spread. Removing realty, the most extreme single sector, still leaves the five session spread wider than all but 1.08 per cent. By twenty sessions 18.6 per cent of windows were wider, and by forty the spread was narrower than the median. The broad index behaved the other way round: an ordinary first day, then five and ten session falls in its worst 1.2 and 2.1 per cent, and a forty session result still in its worst 13.9 per cent. The cross section carried an acute shock that reverted within weeks. The index carried a slower decline, and the slower decline is exactly where the confounds accumulate.

Nor did the spread come from sectors decoupling day by day. The mean pairwise correlation of daily sector returns was 0.55 over the 250 sessions before the announcement and 0.53 over the twenty after. The dispersion came from a few large one directional repricings, realty and public sector banks above all, laid over ordinary co-movement: the opposite of the episodes in stress testing against Indian events, where correlation rose and every sector fell together, and whose screens never selected this episode at all.

A control and a stated window change the answer

A raw return mixes two things: what the whole market did, and what the sector did beyond that. The standard way to separate them is a market model. Estimate each sector's beta against the broad index on a window that ends well before the event, here 247 single session returns from 29 September 2015 to 7 October 2016, ending twenty sessions before the announcement so that the estimate cannot absorb it. The abnormal return is the sector's return minus what that beta predicted from the broad index's move the same day, summed from 9 November.

The control changes the size of the answer and sometimes its reading. Public sector banks rose 1.21 per cent over twenty sessions, which reads as a mild gain. With a beta of 1.62, the broad index's -5.17% predicted -8.37%, so the abnormal return was +10.92%. The banking index fell 6.50 per cent, which reads as damage; its beta predicted -6.03%, and its abnormal return was -0.72%. It did almost exactly what the market move implied.

The window changes the verdict even where the effect does not change. Realty's abnormal return was -14.13% over five sessions and still -8.29% over sixty. The loss barely shrank. What grew is the ordinary variation it has to be seen against: sixty sessions of noise are about 3.5 times five sessions of it, the square root of twelve. The five session figure was more extreme than all but 0.06 per cent of realty's own five session windows since 2013; the sixty session figure was matched or exceeded in 31.1 per cent of its sixty session windows. At sixty sessions no sector's abnormal return was in the most extreme tenth of its own history, the rarest being metals at 14.4 per cent. A study that picks sixty sessions reports nothing unusual; one that picks five reports one of the most violent weeks in the sector's record. Both are right about their own window, which is why the window has to be stated before the result is seen.

Two caveats belong next to that table. The t statistics a textbook would attach use the pre-event residual volatility, which roughly doubled after the event in the sectors that moved most, and daily index returns are not independent, which makes nominal test levels unreliable; so the table ranks each result against the sector's own history instead. That history applies one beta to the whole record, an approximation stated rather than hidden. The general case for choosing the comparison deliberately is made in benchmarking against the right null.

Volatility rose where the shock landed, and the fear gauge priced the average

Annualised realised volatility of the broad index was 11.8 per cent over the sixty sessions before the announcement and 18.6 per cent over the twenty after. For realty it went from 28.0 to 56.3, for automobiles from 18.3 to 33.0 and for consumer staples from 13.0 to 22.3. For energy it fell, from 15.8 to 14.8. The volatility shock was as dispersed as the price shock.

India VIX averaged 14.58 over the sixty sessions before and 17.72 over the twenty after, its highest close in those twenty was 20.15 on 15 November 2016, and its twenty session change from the eve, +0.48%, was exceeded by 44 per cent of twenty session windows. An implied volatility on the broad index prices the broad index, and for the broad index this was a moderate event. The exchange's index files carry a single volatility index and none for the sectors that carried the shock, which is one more reason a VIX zone is a statement about the average and not about any particular book.

Foreign positioning moved before the unscheduled event, not after it

The participant wise open interest file splits index futures positions by category. For the foreign institutional category the net long position, long contracts minus short, was 193,999 on 6 October 2016 and 51,769 on 8 November: it was cut by 142,230 contracts in the twenty sessions before the announcement. Measured as a share of total index futures open interest, that fall of 33.9 points was exceeded in only 3.2 per cent of twenty session windows since 2016. Over the same sessions the client category added 110,250 net contracts, domestic institutions 16,700 and proprietary traders 15,280. The four nets sum to zero every day, which the build checks, because every contract has two sides; the published cash flow figures leave this half of the book out entirely, as the piece on institutional flow regimes sets out.

After the announcement the foreign book did nothing unusual. It added 11,268 net long contracts on 9 November itself. Five sessions later its share of open interest was down 10.6 points, a fall that 13 per cent of five session windows exceeded; twenty sessions later it was down 2.1 points, with 47 per cent of windows showing a larger fall, the middle of the distribution. The brief net short on 24 November is the expiry, when total index futures open interest fell from 381,687 to 267,929 contracts, not a decision.

The timing carries the logic. The announcement was unscheduled, so nothing done before the evening of 8 November can be a response to it. The United States election had a fixed date. The data fit positioning ahead of a scheduled event better than a reaction to an unscheduled one, and they cannot prove either.

What recovered, and what a recovery table cannot tell you

The broad index first closed back at its 8 November level on 25 January 2017, 55 sessions later. Information technology, metals and energy were back within 13 sessions. Every other sector took between 48 and 78, and realty took longest, regaining its level on 1 March 2017. Automobiles and financial services regained theirs on 1 February 2017, the day the Union Budget was presented, when the broad index rose 1.81 per cent. A recovery date that coincides with another event is a recovery the event study cannot claim.

A year on, the table stops describing the event at all. On 9 November 2017 realty, the sector hit hardest, was the best performer of the eleven, up 54.5 per cent from its 8 November 2016 close, and pharmaceuticals, an export sector, was the only one still below it, down 11.5. On 18 September 2026 the media index closed 45.7 per cent below its 8 November 2016 level, and it has been below that level on 85 per cent of sessions since. Nobody would assign a decade of the media index to a currency decision, yet a recovery column invites exactly that reading. After the first few weeks every row of a recovery table measures everything that happened, and the event is one ingredient among many.

An event study is a comparison, so name what it is compared with

Everything above reduces to five rules that apply to any policy announcement, election result or shock.

State the window before looking. The same data called this event extreme at five sessions and unremarkable at sixty. Report several windows, fixed in advance, and say which one carries the claim.

Name the control. A raw return, a market model and the sector's own history gave different sizes, and for public sector banks a different reading. A result without a named control is a description, not a measurement.

Report the spread, not only the average. The broad index averaged a realty collapse and a public sector bank rally into a 5 per cent week. A book is held in its own sectors, not in the average of all of them, which is the same point sector rotation makes about leadership.

Check the calendar for rival events. A same day confound cannot be removed by statistics. Look for a market where the rival explanations predict opposite signs, as government bonds did here, and accept that elsewhere the split is unknown.

Distrust recovery tables beyond the event window. After a few weeks the column measures the following year's economy and policy, not the event.

What the measurement is for

The practical use is in how a book is stress tested and how a record is read. A book concentrated in realty and consumer sectors lived through a twenty per cent shock that the benchmark recorded as a five per cent week, and a scenario library screened on the broad index alone does not contain it. A pooled performance figure silently averages over episodes like this one, the argument of regime conditional performance, and a flow headline read without the derivative book misses what foreign investors did, and when. Reading an event through the cross section and a stated control, rather than through the index and a headline, is a method, and it is taught as one.