Abstract

Up to November 2026 we ran about 38 pre-registered rounds of intraday tests on NSE single-stock F&O underlyings (numbered up to 42; several were dropped or never run), covering cross-sectional reversal and momentum, opening-range breakouts with various stock screens, corporate-announcement and news events, index rebalancing, forced-flow and positioning signals, volatility-compression setups and execution variants. Most rounds ended at the first (development) gate. A few effects were confirmed out of sample, but almost every intraday effect we found was small relative to a realistic round-trip cost of about 10.5 basis points. The most consistent structural result was that information we could predict before the open was already reflected in the opening gap, leaving little for the trading session. One further finding — that volatility compression predicts quieter days rather than breakouts — contradicts a widely repeated practitioner claim.

Data

  • One-minute bars for over 200 NSE F&O stock underlyings, 2020 onwards, split- and bonus-adjusted.
  • NSE corporate announcements, delivery-percentage data, F&O bhavcopy (including open interest and market-wide position limits), index-rebalancing records and India VIX.
  • A cost model with statutory charges (securities transaction tax, exchange and regulator fees, GST, stamp duty) calibrated to real contract notes, plus a per-stock spread estimate. The cost gate for an intraday cash round trip was about 10.5 bps (statutory charges plus a fixed spread allowance), of which roughly 4.5 bps is statutory and cannot be avoided by better execution.

Design

Every round followed the same protocol:

  1. Pre-registration. The rule, the data splits, the cost assumption and the pass criteria were written down and committed before any returns were examined.
  2. Step 0 feasibility checks. Before computing any P&L: are there enough events in each split, and is the average move large enough to plausibly clear costs? For signals available before the open, does the effect show up in the session or only in the opening gap?
  3. Sequential splits. Development 2020–22, validation 2023–24, test 2025 (April–December), and a 2026 holdout that may be opened at most once per rule. A rule that fails a gate does not see the next split.
  4. Red-team checks. Drop the top 10% of trades, compare against placebo days and matched controls, and double the cost assumption.
  5. Multiple-testing control. When rounds were run in batches, the Benjamini–Hochberg procedure was applied across the batch.

From round 24 onward we added a written standalone-profitability standard: a rule passes only if its net result has a confidence interval above zero at a conservative cost level, it trades often enough to matter, it survives dropping its best trades, every year is positive, it meets minimum return and maximum drawdown thresholds, and it beats its control. "Statistically real but not tradable" does not count as a pass.

Findings

1. Costs dominate

The recurring pattern across rounds was an effect that existed but was worth only a few basis points. Examples:

  • Opening-range breakouts on "stocks in play" earned about +6 bps gross per trade, stable across every year in development and validation, and negative after realistic costs.
  • A sweep of 144 intraday configurations (opening-range, gap, volume-shock, late-day momentum and VWAP variants) found the best ones grossing around 13 bps against roughly 12 bps of realistic cost.
  • Late-day continuation of strong movers (see finding 3) grossed about 7 bps against 10.5 bps of cost.

At this cost level, an intraday rule needs a gross edge well above 10 bps per trade to have room for slippage, bad fills and decay. Very few mechanisms produced that.

2. Predictable information is priced in the overnight gap

Several unrelated signal families showed the same shape: whatever could be predicted from information available before 09:15 appeared in the move from the previous close to the open, and the session itself was flat or slightly reversing.

  • On F&O ban-entry days, roughly 21 bps of the stock's underperformance sat in the opening gap, which cannot be captured by a trade entered at the open. (The session decline that followed is a separate result, covered in the F&O ban paper.)
  • Signals built from options positioning, and option-led direction on unscheduled-news days, predicted the gap but saw the session reverse by around 12 bps.
  • Up-gaps on buyback and order-win announcements opened higher and then faded through the session.

This became a standing rule: any pre-open signal must pass a gap-versus-session check before it is tested as an intraday strategy.

3. One effect confirmed out of sample — and still not tradable

Afternoon continuation (13:31 to 15:14) of strong movers in speculation-heavy stocks — those with low delivery percentages — beat random entries, low-speculation stocks and the full universe, with confidence intervals above zero in the 2023–24 validation period. It is the clearest example of a real intraday behavioural effect in the programme.

It still failed. Gross profit was about 6.9 bps against 10.5 bps of cost, giving a net result of about −3.6 bps, and the effect shrank over time, from roughly +10.6 bps net in 2020 to −6.5 bps net in 2024.

4. Compression predicts quiet, not expansion

A popular practitioner claim is that narrow-range days (NR7) and inside days precede range expansion and breakouts. The data said the opposite. The day after a compressed day had a range of about 0.92 times its 20-day average, against about 1.03 times on other days (difference about −0.11, confidence interval −0.13 to −0.09). Opening-range breakouts on the day after compression did not do significantly better than the same breakouts on ordinary days.

5. A result that passed validation and was killed by the red team

A pre-registered fade of large up-gaps on buyback and order-win announcements passed all four validation gates in 2023–24, at about +52 bps net per trade with a confidence interval above zero, and beat same-size gaps without news. Dropping its ten best trades turned it to about −16 bps. With only 20–40 qualifying events a year and a hypothesis generated from earlier development data, it was closed.

6. Execution changes do not rescue a weak signal

Four rounds tested passive limit entries and stop or confirmation entries on the two most promising rules. Limit orders filled more often on trades that went wrong and missed the runners; stop entries did the reverse. In both cases a better entry price was traded away for worse trade selection, and the two cancelled out.

7. Null results

The following produced no tradable effect at the development or validation gate: 30-minute cross-sectional reversal; weekly residual reversal; screened and unscreened opening-range breakouts; index-rebalancing adds and deletes; broad fades of non-results news gaps; fading sector peers that gap in sympathy with a news-driven leader; reversal on high-stress index days (prices continued rather than reversed); fading strong movers before the broker square-off window; morning-strength continuation; strike pinning at monthly expiry; price-band "magnet" effects; and slow-horizon volume factors.

A data lesson worth repeating

Event studies depend on knowing when information became public. Partway through the programme we found that the timestamps attached to some corporate filings may mark a later filing event rather than the first public release: most carried timestamps after the 15:30 close. A rule tested on such timestamps may not be testing what it claims to test. Affected results are treated as provisional until event timing is rebuilt from first-release announcement times.

Limitations

  • Results apply to NSE F&O stock underlyings traded intraday in cash, at retail cost levels. Lower costs would change several verdicts.
  • Spreads are estimated from bar data rather than measured quotes for most of the sample.
  • Rare-event rules have small samples even over six years, and a single-year test split is a weak gate for them.
  • Across the programme many hypotheses were tested; individual near-misses should be read in that light.
  • One rule passed its 2026 holdout gate, but fragilely and on event timestamps now known to be provisional; it is not reported here as a finding.

Conclusion

Intraday edges in liquid Indian single stocks exist, but on this evidence they are mostly smaller than what it costs a retail participant to trade them. The effects that did hold up across our wider research were structural — tied to market rules or to the price of options insurance — rather than intraday price patterns. The current direction of this work is a point-in-time event atlas that maps which event types could plausibly support a standalone strategy before any returns are examined.