Posted in My Daily Journal · 9 July 2026
A systematic NIFTY pivot system went live on futures this week. We used the first days to answer the questions a live position forces on you — is stop-and-reverse right, should big open profits be booked early — and found that a modelled option overlay was four times too optimistic.
What is the system, in one paragraph?
NIFTY 10-minute bars, fractal pivots confirmed five bars later, a long when price breaks the last pivot high and a short when it breaks the last pivot low, a skip when the box between the two pivots is under 0.30% of the index, and a stop that ratchets one way to the latest opposite pivot. It reverses in the same bar when the stop is also the opposite signal, holds overnight, and stays in from 09:16 to 15:25. One futures lot, fixed. Nothing about it is tuned to a year.
On the first real-money day every order was rejected for twenty minutes. What happened?
Market orders with a broker-side "market protection" build a synthetic limit a few percent away, and that limit sat outside the exchange's price band for index futures. Paper trading could never have found it because paper places no orders. The fix was a marketable limit order a small fraction from the touch; no money was lost, and the lesson is permanent: the only test of an order path is a real order.
A long was up ₹7.5k during the day, booked only ₹1k, then flipped short. Is stop-and-reverse right?
Yes — the flips are the strongest part of the system. Replaying the exact live configuration over 872 trades, 580 were flip entries and they produced ₹5.0L of the ₹7.33L total, averaging ₹863 per trade against ₹797 for fresh signals. The variant that exits and waits for a fresh pivot made ₹1.75L less with a worse profit-to-drawdown ratio. Giving back part of a move and flipping is priced into the system's payoff; removing it removes the engine.
Why not book big profits early and re-enter later?
Because it is monotonically destructive. A ₹600 take-profit cost ₹1.35L of the four-year ₹7.33L; ₹400 cost ₹3.2L; ₹200 left ₹1.42L with a worse drawdown. The win rate rises from 38% to 41% while the profit collapses — the psychological trap, measured. Drawdowns did not improve at any level because drawdowns come from losing streaks, not from winners giving back. Forty precedent trades at the same open profit as the day's position showed that holding averaged ₹1,995 better per trade, with five fat tails adding ₹25–38k each.
A Black-Scholes model said buying weekly ATM options on the same signals would have turned ₹1L into ₹3L in six months. Did real premiums agree?
No. On real NSE premiums the same signals finished at ₹1.48L with a ₹1.03L drawdown — ruin-grade on ₹1L — and the worst plausible fills went bust in seventeen trades. The model was about four times too optimistic, and daily option data could not even price the 70% of trades that happen intraday. For weekly ATM buying, fill quality and timing dominate the strategy itself; futures do not have this pathology, and their live fills matched the backtest within rupees. The overlay was shelved pending minute-level option data.
Any engineering lessons from auditing the live bot?
Four, all of the same kind — the code assumed orders are all-or-nothing at exactly one lot. Partial fills were invisible, so a cancelled order could leave a position the bot did not know about and the next exit could reverse exposure by mistake; rolls assumed lot sizes never change between expiries (the exchange has changed them); state writes were not atomic; externally closed trades never reached the journal. Every one was fixed before the next contract roll. A trading bot's hardest bugs are quantity bugs.