Can a NIFTY option buyer trade commodities? Rolls, real option prices, and an intraday engine that fails

Three things, in order: the deep in-the-money option must actually be tradeable (a spread you can cross and depth for one lot), the futures backtest must survive a proper model of contract rolls, and the option must capture most of the futures move after decay, rolls and costs. None of these can be assumed from a futures backtest; each needs its own measurement. We started all three last night and finished the second and third today. The liquidity measurement needs ten trading days before it says anything.

You record it yourself, live. Every fifteen minutes through the whole MCX session a read-only logger snapshots the full quote — best bid and ask with sizes, five levels of depth, last trade time, volume, open interest — for every strike within ten percent of the futures price on the two nearest expiries, for gold, silver and their mini contracts. From ten days of that you get the half-spread as a percent of premium by moneyness band and by hour of the session, the share of snapshots with a lot available at the touch, and how many days each strike actually traded. The pass bars were written before the first snapshot: median half-spread at or under half a percent of premium at two percent or more in the money, depth for a lot in nine snapshots out of ten, traded on four days out of five. A single evening snapshot already reversed our prediction: the big gold and silver contracts' options are dead — half-spreads of ten to twenty percent of premium — while the mini contracts' options at two to six percent in the money quote at a third to half a percent, with a lot on both sides. Only the nearest monthly expiry trades; the second expiry is one-sided or empty everywhere. One snapshot is not a verdict. Ten days of them will be.

The roll. When the near contract expires, the series steps up into the next contract, and on metals that step is the carry: measured directly, the first trading day after expiry shows a median overnight gap of 0.3% on gold mini (monthly), 0.7% on the big gold contract, and 1.7–1.9% on silver and silver mini (every two to three months), against roughly zero on every other day. That is five to ten percent a year. A long position held across a roll gets that step credited in a continuous-series backtest and pays it in real life. Charging it two ways — the measured step per roll crossed, and this year's observed calendar spread as a daily carry — removed 16 to 39 percent of last week's profit, and turned the 2021–23 half negative on the big silver contract and, under the harsher model, on gold mini. The only name that passed cleanly was silver mini, which rolls least often.

Yes, once we found that the vendor still serves expired contracts if you ask by the contract's own name. That gave every expired gold, gold-mini, silver and silver-mini future since late 2018 — 48, 95, 40 and 35 contracts — minute by minute for each contract's last three months. Rebuilding the series contract by contract, rolling at the close of the first day the next contract out-trades the current one, with signals on a back-adjusted series and P&L on actual contract prices, the common window with last week's test came out: gold ₹51.9L per lot (82% of the continuous-series figure), silver ₹48.5L (81%, but its early half is negative — fails), silver mini ₹9.4L (83%, passes, early half exactly zero), and gold mini ₹6.4L — 110%, more than the continuous series showed. The continuous series had been carrying junk bars around roll periods that whipsawed the gold-mini signal; on real contracts its 2021–23 half is positive and it is the least bull-dependent of the four, the only one that earned in 2022, 2023 and 2024. The two and a half years before 2021 that the earlier test never saw were positive on all four, nearly all of it 2020. Flat years cost little — the worst year on silver mini was −₹15k — so the structure is a trend-capture with a small bleed, not a loser, when metals do not trend. Both minis are the candidates; both big contracts are out.

With daily closes of every call on the minis since late 2020 (gold) and mid-2021 (silver), the rule was: at each long signal buy the nearest-expiry call closest to three percent in the money, roll it three days before expiry, sell it when the signal goes flat, one lot, exchange costs on every leg. Gold mini: 49 of 67 signals priced (the 18 missing are mostly 2021–22, when those options barely traded), futures ₹6.0L against option ₹5.7L — a capture of 0.95, and never below 0.77 in any year since 2023. Silver mini: 55 of 61 priced, futures ₹11.4L against option ₹10.5L, capture 0.92 — but in the small years, 2023 and 2024, decay and rolls ate most of a ₹35–40k futures year (capture 0.63 and 0.25). The option's drawdown runs 45–50% deeper than the future's, about ₹1.0–1.2L per lot over the sample, against a median entry premium of ₹17–23k. Against the bars written beforehand, gold mini passed tracking and data coverage and failed the drawdown bar; silver mini failed tracking in its flat years and the drawdown bar. That is physics, not a flaw: a long option bleeds when nothing happens. For a buyer-only book the comparison is not option versus futures but option versus nothing, and the option keeps 92–95% of the profit overall and 95–102% in the years that paid.

With great suspicion. The first pass had the gold-mini option out-earning the future it tracks, which is impossible for a near-delta-one instrument. The cause was a single one-lot print early on a day gold fell twelve percent: by the close the call was out of the money, but the day's "close" was that morning print, and the next morning it opened at a fifth of the price. Flooring at intrinsic value does nothing against an inflated stale print. The rule that held: mark every leg at intrinsic value against the day's futures close plus the time value seen at the option's last print, and never credit more time value at exit than was paid at entry. Conservative for a long option, and it removed the phantom profit in both directions. Every correction was logged against the same data before the final numbers were taken, with the pass bars unchanged.

No, anywhere. The exact engine — fractal pivots, 0.30% box, opposite-pivot stop, two lots with one booked at 2.33 boxes, overnight carry — ran on all four metals over eight years of real contracts, with entries allowed in four pre-registered windows: the whole session, the day session to five, the evening US session, and the whole session minus the US data hour. Sixteen cells, sixteen failures of the same bar: the 2018–2022 half is negative in every one, and essentially every rupee of profit is the 2026 silver run — more than the whole total on silver mini. Gold and gold mini lost in every year from 2019 to 2024 in every window. Our prediction that silver's evening session would pass was wrong; the evening is no better than the day. It is the same shape we found on NIFTY before 2020: the pivot structure is a high-volatility-regime engine, and metals were not that regime until 2025. Closed, with no confirmatory run and no crude run; a different bar or box would be a new question, and it would have to earn in 2018–2024 first.

One lead, two names, no book yet. The daily 5-EMA stop-and-reverse on the silver-mini and gold-mini contracts, expressed as a three-percent in-the-money call, with the December decision to be written on the option numbers — ₹10.5L and ₹5.7L per lot over five to six years with ₹1–1.2L drawdowns — not on the futures line. Two inputs are still accumulating unattended: the ten-day liquidity table, and a paper journal that buys and sells the option at the quoted ask and bid of the morning snapshot and records how much of the futures shadow each closed trade captures; the backtest says about 0.9, and a sustained figure under 0.7 on trending trades would mean live fills are worse than the backtest's marks. Fragility stated once: 80% of the silver-mini option total is a single trade from last October to January, and 2025–26 is 98% of it; gold mini is the steadier tracker, silver mini the bigger earner.