Room 07 · Laboratory · data to 15 Oct 2026
The laboratory.History, tested.
Each experiment asks what followed a reading in the past, using only what was known on the day, and sets the answer against what followed every other day. Results that could be luck are called luck. Nothing here predicts anything.
How many times the reading happened. Cases whose windows overlap share their outcome, so they count once as independent.
The baseline: what followed every eligible day over the same window. A case is only interesting if it differs from this.
How often random sets of days, as many as the cases, did as well or as badly: 10,000 seeded draws, so the result is reproducible.
A fixed rule. Under five independent cases: too few. A two-sided chance check under 5%: unusual. Otherwise: within the range of chance.
Each dot is one fall of 10% from the record, and what the Nifty 50 did over the next six months. There are 7, each taken on the day it happened, with nothing from later.
After the Nifty 50 first falls 10% below its record
Two stories pull opposite ways. Falls can feed on themselves, as volatility clusters; or a deep fall can leave prices that later recover. If either showed up, the cases should sit apart from every other day.
Each fall counts once: the first close 10% or more below the record close so far, re-armed only by a new record. Outcome: the Nifty 50's change over the next 6 months. Baseline: the same change after every day since Oct 2016.
Grey bars: what followed every day since Oct 2016 (2,341 days). White dots: the 7 finished cases.
+5.3% median over 6 months; 5 of 7 higher.
In 44% of 10,000 random draws of 7 days, the median was as low as this. Random days do this often, so the difference is within the range of chance.
| Day | After |
|---|---|
| 12 Mar 2026 | −1.8% |
| 13 Nov 2024 | +5.3% |
| 20 Dec 2021 | −7.2% |
| 6 Mar 2020 | +2.6% |
| 5 Aug 2019 | +11.5% |
| 5 Oct 2018 | +12.9% |
| 23 Mar 2018 | +10.6% |
“Open”: the window has not finished yet; open cases are not counted.
After the first close below the 200-day average
Trend followers treat the 200-day line as a boundary: below it, falls are said to run further. If that held here, the cases should lag every other day.
A case is the first close below the 200-day average after at least 20 closes above it. The average is of the 200 closes before the day, so it was known in advance. Outcome: the change over the next 3 months; the table adds the sessions until the first close back above.
Grey bars: what followed every day since Aug 2017 (2,204 days). White dots: the 11 finished cases.
+1.9% median over 3 months; 6 of 11 higher.
In 29% of 10,000 random draws of 11 days, the median was as low as this. Random days do this often, so the difference is within the range of chance.
| Day | After | Back above |
|---|---|---|
| 23 Jan 2026 | −4.2% | 1 session |
| 20 Dec 2024 | −1.0% | 8 sessions |
| 14 Nov 2024 | −2.1% | 4 sessions |
| 28 Feb 2023 | +7.5% | 1 session |
| 28 Sep 2022 | +7.9% | 2 sessions |
| 19 Apr 2022 | −4.0% | 2 sessions |
| 24 Feb 2022 | +2.1% | 14 sessions |
| 26 Feb 2020 | −13.8% | 95 sessions |
| 30 Jul 2019 | +7.9% | 34 sessions |
| 4 Oct 2018 | +1.9% | 30 sessions |
| 19 Mar 2018 | +6.1% | 6 sessions |
“Open”: the window has not finished yet; open cases are not counted.
Is high fear followed by bigger moves?
India VIX prices the move options traders expect over the next month. If that expectation carries information, fearful readings should be followed by bigger moves, in either direction.
Every 21 sessions since 19 Oct 2022, so no two windows overlap: India VIX's percentile among its closes so far, then the size of the Nifty 50's move over the next 21 sessions. Fearful means at or above the 70th percentile, the fear condition's line. Chance check: the labels shuffled 10,000 times.
Each dot is one window: the size of the move, up or down. The white tick is each group's median.
Fearful windows moved 2.2 points more at the median. Shuffled labels gave a gap at least that wide, either way, 8% of the time, so it is within the range of chance.
Is any month of the year different?
Calendar effects (budget months, festival seasons, results seasons) are quoted every year. With a decade of data, each month has about ten observations, which is very little to go on.
Month returns from month-end closes, 2016–2026; the month in progress is left out. Each month's median is set against the medians of as many months drawn at random from all 119 of them.
One dot per year; the white tick is the month's median; the dashed line is the median of every month (+1.1%). Hollow: off the scale (Mar 2020 −23.3%).
No month's median stands apart from months drawn at random. With about ten years per month, the calendar shows no pattern this data can separate from luck.
Do leading sectors keep leading?
Momentum says recent winners keep winning for a while; short-term reversal says the opposite over a few weeks. Ranking sectors lets both be checked at once.
Every four weeks since 17 Jan 2025: rank 22 sectors by their past 13 weeks, then compare with their next 4 weeks (rank correlation, from −1 to +1). Sectors are equal-weight indices of today's Nifty 500 members. Chance check: next returns shuffled among sectors within each period, 5,000 times.
Filled bars: periods where past leaders also led next; hollow bars: periods where they lagged.
Past leaders tended to keep leading over the next four weeks. Shuffled returns gave an average at least that far from zero 45% of the time, so it is within the range of chance.
After breadth slides below 30%
Washed-out breadth is often called a sign of capitulation. The breadth record here starts in Aug 2025, so this experiment mostly shows how little history there is to test the idea on.
A case is the first close with less than 30% of the 2200 stocks above their 200-day average, after breadth had been at 40% or more. Today's index members, applied to the past.
Grey bars: what followed every day since Aug 2025 (275 days). White dots: the 1 finished case.
+1.1% median over 1 month; 1 of 1 higher.
1 independent case: too few to tell a pattern from luck. The cases are listed as a record, not as evidence.
| Day | After | Breadth |
|---|---|---|
| 9 Jan 2026 | +1.1% | 28% |
“Open”: the window has not finished yet; open cases are not counted.
Every result is rebuilt from the collected data each night. Verdicts follow the rule above: too few cases, unusual against chance, within the range of chance. Past patterns do not predict future returns. Method and sources