Posted in My Daily Journal · 10 July 2026
The full 2015–2026 stress test of the live configuration on 106,956 ten-minute bars, why a volatility filter was tested and rejected, and how the stand-down policy was rewritten so it would not have quit right before the best era.
What did the full-history test show?
On 2,730 trades with the live configuration and real costs: net +₹11.88L, win rate 37.5%, payoff 1.94, maximum drawdown ₹2.50L, profit-to-drawdown 4.75. By year: 2015 +36k, 2016 −65k, 2017 −69k, 2018 −27k, then eight straight winning years from 2019 to 2026 including COVID (+148k in 2020) and +311k in 2022. The edge is real across regimes, and it starved for three consecutive years.
Three losing years in a row — would the stand-down rule have stopped it?
It would have fired in about 2017, and the system would have missed the golden era that followed. The old rule was "three losing months with the cumulative loss below a threshold", and the 2016–18 grind trips it. That is the real danger of a stand-down rule: it is a forecast dressed as risk control. We replaced it with a rule that only quits on something out of distribution — a hard drawdown stop well beyond the historical maximum, a statistical stop on a rolling twelve-month loss far below any observed, an immediate kill-switch on mechanical anomalies, and an annual review against the backtest envelope. The principle written down: reduce size, don't tighten triggers.
Would a volatility filter have avoided the dead years?
No. Gating entries by an ATR-percentage threshold at every level from 0.7 to 1.1 cut the 2016–18 bleed only by taxing the good years more — totals fell by ₹1.2L to ₹5.5L — and the maximum drawdown stayed at ₹2.44L at every setting because the deep drawdowns happen in volatile chop, not quiet markets. Losing years simply migrated to different years. The winters were bad trend quality, not low volatility, and the question was closed in both directions.
Is the maximum drawdown survivable on a small account?
The full-history ₹2.50L exceeds the four-year window's ₹1.81L, which is the usual lesson of looking further back. The capital rule that came out of it: fund a lot with its premium or margin plus the measured maximum drawdown, and treat a front-loaded losing sequence as the base case, not the exception.