Posted in My Daily Journal · 27 October 2026
We re-ran the live intraday chain end to end on a new option archive, hunted for ways to make winners bigger without making losers bigger, tested two indices and 117 stocks on a popular breakout toolkit, and pre-registered an overnight-continuation idea that failed its bar.
Does the gap-and-go chain survive being replayed on a completely different option data set?
Yes, with the same shape. Signals from the new feed's index minutes and legs from its expired-option files reproduced the decision on 99.4% of 1,428 sessions, the same entry minute on 95% of trades and the same exit reason on 97%. Totals: ₹2.94L versus ₹3.20L per lot over 2021–Oct 2026 — the 8% gap is option-leg prints at stop and trail levels, concentrated in 2025–26. The edge in 2021–24 is robust to the data source; 2025–26 is small on every feed (₹441–1,156 per trade at five lots), with 2023 negative on both.
Can we make the winners bigger and the losers smaller, whatever happens to the win rate?
The payoff is already "few big wins": average win ₹3,911, average loss −₹2,086, win rate 41%, and the ten biggest wins are 64% of six years' profit. On a strike ladder, deeper in-the-money bought bigger wins only with bigger losers and a worse profit-to-drawdown ratio; out-of-the-money could not deliver bigger wins on weeklies because of time decay. The only lever that made wins bigger without making losers bigger was holding a 15:15 winner overnight — and it did so by accepting gap risk. Nothing new on the loss side.
What did the overnight hold look like once pre-registered properly?
It failed its bar. Holding a chain trade that was up 20% or more at 15:15 to the next morning added ₹1,100 per trade on 122 cases, but the confidence interval included zero, the worst hold (−₹17,109) exceeded three times the average win, and the out-of-sample year added only ₹50. All of the edge was Friday to Monday: 30 holds at ₹4,532 each, 83% winners — too few to adopt, and two of seven years negative. It became a forward paper log of twenty Friday holds, not a rule. A first version of the results had analysed every 15:15 exit without the 20% gate; the correction was recorded, not hidden.
BankNifty is a different index. Does the chain work there on real monthly-contract prices?
No. On 456 sessions of real prices (Nov 2024 → Sep 2026) every variant loses: the door alone −₹263 per trade at a 0.5% spread, the chained version −₹419, Williams standalone −₹461. The earlier model had been ₹10–15k a year too kind. Two independent researchers then tried 50 more BankNifty buying ideas on the same data: zero passed. The structural reason is that a daily at-the-money straddle on BankNifty monthlies loses about ₹348 a day gross — time decay exceeds realised movement — so every directional buy starts from a negative carry.
A popular post says 15-minute opening-range breaks with displacement candles and fair-value gaps are the way to trade indices. Is it?
Tested without bias on NIFTY (real ITM legs, 2015–2026), BankNifty (real monthly options) and 117 stocks: no cell passed and nothing beat our plain gap rule by ₹150 per trade in both halves. The reasons are instructive. A raw 15-minute breakout fires on 99% of sessions, so it selects nothing. A fair-value gap was present in 99.9% of NIFTY breakout windows — as a confirmation it filters nothing. Displacement candles hurt by about ₹500 per trade. Liquidity-target exits won 24–27% of the time with negative profit-to-drawdown. The one real idea in the post — directional bias from the opening gap — is already the whole of our gap-and-go.
Anything positive from the day?
One structural pattern appeared on both indices from independent code: overnight gap continuation after a trend day — a 15:15 winner on NIFTY, an afternoon breakout on BankNifty — earns ₹700–1,350 a trade, almost all of it on Fridays, with full-premium gap risk and five-trade concentration. Real, thin, and not yet a system. It entered the research queue as the next paper candidate, pre-registration first.