Posted in My Daily Journal · 12 June 2026
The starting point of this journal — a blunt summary of what testing more than thirty retail strategies on Indian equities showed, the one idea that survived, and the launch of a NIFTY pivot-breakout system in paper mode.
After testing more than thirty retail strategies on liquid Indian stocks, what actually survived?
Almost nothing intraday. Every intraday archetype we could find — fading, breakouts, scalping, moving-average crosses, Heikin-Ashi, opening-range breakouts, top-mover chasing, 5-EMA touches, Bollinger fakeouts — lost after the roughly 14-basis-point round-trip cost floor and whipsaw. Buy-and-hold beat every active strategy on the NIFTY 50 basket. The one durable edge was multi-day small-cap momentum convexity (breakout to a 100-day high above the 200-day average, regime-gated, chandelier trail, let winners run): payoff about 3.8 to 1 with a fat tail up to +73R, but survivorship-inflated, regime-dependent (great in 2020, 2021 and 2023; negative in 2019, 2025 and 2026) and needing months-long holds.
Did any indicator ever improve a strategy rather than just decorate it?
One. ADX(14) above 25 on the daily timeframe tripled the expectancy of a 20-day-average pullback entry — the only filter that improved anything. The same filter was useless intraday. Shorting equities for bearish exposure lost in three separate tests; the bearish idea that held up was buying puts in a down regime near the 20-day average with VIX below its 80th percentile — tail-dependent and sample-fragile, but convex in the right direction.
Why build a pivot-breakout system on the NIFTY future rather than another indicator strategy?
Because it came from a verified brokerage-contest winner's published rules and it passed every robustness gate we had: NIFTY 10-minute fractal pivots (five bars either side), breakout above the last pivot high or below the last pivot low, no profit target, a ratcheting stop at the opposite pivot, overnight holds, both directions. On four years of 10-minute index data it was the first strategy out of about fifty to pass all gates — roughly ₹4L per lot over four years on pessimistic fills, positive every year. It went live in paper mode with the warning on record that a 33% win rate and rare large winners mean long losing stretches by design.
Was the "convexity principle" — cut losers, let winners run — a universal improvement?
No, and that was the useful finding. Chandelier-style exits built real convexity (tails to +41R) but only paid when both the entry and the instrument trend. On all four large-cap intraday setups we tested, letting it run was worse than capping the trade intraday. Convexity has a home: multi-day positions in things that trend. Applied to intraday noise it just widens the losses.
You also ran an LLM "persona swarm" that called the market up or down twice a day. How did it do?
In its first weeks it looked brilliant — five wins out of six settled calls. It was also calling UP every single time during a rally. The only loss was the one down day. That is bull-market beta, not forecasting skill, and it was labelled an experiment, never to be traded. A proper out-of-sample test of the same idea came later in the year and found no signal.