Posted in My Daily Journal · 9 October 2026
A long day of pre-registered tests on second trades after a loss — every re-entry rule, every market-reading policy — ending with a complementary breakout rule and a routing rule that passed every bar.
If the gap-and-go buyer gets stopped out, can it re-enter later in the day on a specific event?
No rule clears the bar. Of 1,256 first trades, 471 were losses before 14:29 — there is time for a second trade — so we tested re-breaks with retest, re-crosses, new highs of the day, reversals through the opposite range boundary, stop-and-reverse at the exit, cool-off periods and fixed times. Every same-side re-entry lost: −₹461 per trade on re-breaks, −₹326 on re-crosses, −₹281 on new highs. A stop at −30% on an in-the-money option means the spot went back through the range; by a second break hours of theta are gone and the day is choppy. Prediction confirmed; nothing was adopted.
One footnote looked promising — reversing on expiry days after a loss made +₹30k. Did it survive scrutiny?
It failed four of five pre-registered tests. Walk-forward out of sample: P/DD 0.8. Mechanism in the spot: the reversal-through-the-opposite-boundary rate was 35% on expiry days versus 38% on other days — no expiry-specific reversal exists. BankNifty: all twelve equivalent cells negative. Luck: random 97-trade slices of the same loss days reached +₹30k about one time in four to six. Only the parameter plateau passed, and a plateau on one small sample measures parameter sensitivity, not whether the sample was lucky. Footnote closed.
What about "disowning all bias" — at the loss exit, read the market and pick the best strategy from a menu?
Every market-reading policy lost out of sample on the 408 loss exits from 2019 to 2026, and the combined book's profit-to-drawdown fell from 4.8 to between 1.8 and 3.5 in every case. A random pick lost ₹66 per trade. An oracle that knew the best choice each day would have made ₹3,029 a day — the information exists in hindsight and no policy recovers any of it. Two contest-winner rules did stand out as standalone trend-day entries rather than after-loss fixes: new day extreme beyond yesterday's high or low (₹417 per trade, 10 of 12 years), and the Williams volatility breakout — spot beyond the open by half of yesterday's range (₹221 per trade, 10 of 12 years).
Where does the Williams breakout fit with gap-and-go?
As its complement, on the days gap-and-go deliberately stands aside. Gap-and-go only trades breakouts aligned with the gap; when the opposite side of the opening range breaks first it stands down. On exactly those days the Williams rule — a counter-gap move of half of yesterday's range — earned ₹181 per trade over 916 trades, positive in both halves of the sample, and the two together reached ₹4.88L with profit-to-drawdown 5.9 versus 5.6 alone. Mechanism in plain words: gap-and-go catches gap-and-go days; Williams catches gap-and-fail days, which become trend days in the other direction.
You then tested the pair properly, including routing by the size of the gap. What came out?
A rule we called GW2: gap under 0.2% → Williams only; 0.2–0.5% → gap-and-go only; 0.5% or more → gap-and-go first, with Williams armed as a backup the minute gap-and-go stands down. One trade a day, never a second after a loss. Twelve years: 1,662 trades (about 2.7 a week), +₹5.80L, ₹348 per trade, 44% winners, maximum drawdown ₹64k, profit-to-drawdown 9.1, positive in 10 of 12 years, both halves positive, walk-forward in both directions, robust to a two-minute fill lag. The routing learned on either half of the data gave the same picks. Caveats on record: the edge concentrates in trend-rich years, 2025 was worse than gap-and-go alone, and Williams' own k of 0.5 was kept rather than tuned because tuning it failed walk-forward.
What did Larry Williams himself say about the rule you borrowed?
In his published work the breakout is today's open plus or minus a fraction of yesterday's range — "just yesterday's range works wonders" — with the fraction tuned per market (50% for the S&P), stops either in money or at half of yesterday's range, and exits at the first profitable open rather than targets, which he found hurt. He also said, decades later, that the original version has not worked as well in recent years. Our data agreed with him on three points: bigger fractions give fewer, higher-quality trades; his contrarian "sell after up days" filter hurt on NIFTY; and no target beats any target.