← Option pricing models & outliers · Research, not advice. Read-only; no orders.

Top Nifty option mismatches vs fair value (1-minute, Sep 2024 – Oct 2026)

Research, not advice. Read-only study on historical data; no orders were placed. Nothing here is a recommendation to trade.

Short answer

• We compared every Nifty weekly option price, every minute, with a "fair" price from a smooth SVI volatility smile fitted to the whole chain that minute. A mismatch is a minute when one option's last traded price sits clearly away from that fair value.
• 1,175,263 raw mismatch episodes → 267,393 liquid ones (option actually traded, price moving) → only 4,055 tradable ones (also confirmed by the put/call partner at the same strike). Most raw mismatches are stale quotes, not mispricings.
• Tradable mismatches are small: median 1.6 pts, 90th percentile 6.0 pts, largest 30.5 pts. 1,063 are ≥5 pts.
• They cluster on expiry days (19.9 per expiry session vs 5.0 on other days), in the first 30 minutes, and right after sharp spot moves (≥12-pt 1-minute move: 33.8% of big tradable episodes vs 9.4% of all minutes, 3.6× more often).
• Big tradable gaps close fast: the median time for a gap to halve is 4 minutes. In 53% the market price moved back to the model; in 38% the model moved to the market.
• Buyer test: FAIL. Buying "cheap" mismatches lost money after costs in the 2026 test (liquid: lost ₹39/trade over 514 trades; tradable: lost ₹182/trade over 303 trades).

What a "mismatch" is

Each minute, for the front and next weekly expiry, we fit one smooth SVI smile through the chain's implied volatilities (OTM strikes, vega-weighted). Put that smile's volatility back into Black-Scholes and you get a fair value for every strike. The gap = market price − fair value, in premium points, in % of fair, and in volatility points. A minute is flagged when the gap is unusual for that contract (robust z-score ≥4 against its own last 60 minutes, and ≥1 pt), or simply big (≥5 pts, ≥20% of fair, or ≥5 vol points). Back-to-back flagged minutes (≤3 min apart, same contract, same direction) count as one episode.

Cheap = market below fair; rich = market above fair. Costs used everywhere: lot 65, 1 pt slippage each side, ₹50 fees per round trip.

Raw vs liquid vs tradable, and why most raw ones are stale

At the minute level, of 88,453 strong raw outlier minutes (|z|≥4, ≥1 pt): 67.4% had no trade that minute (the "price" is the last old trade), 52.3% had no trade the next minute either, 48.1% showed a price unchanged for 3+ minutes, and 98.5% were not confirmed by the put/call partner. Only 1,106 (1.3%) passed every check. So a raw "mismatch" is usually a stale last-trade price while spot and the rest of the chain moved on, not a price anyone could actually trade at.

Liquid-but-unconfirmed gaps are mostly deep in-the-money strikes (99% of liquid ≥5-pt episodes are ITM; 58% are 500–1000 pts from spot). These close within one minute (84.6% halve in ≤1 min) — typical of thin, wide bid/ask prints bouncing, not a real mispricing.

Top 10 tradable mismatches (by premium points)

"Volume that min" is in contracts (units), as in the source data. Full top-200 lists by points, %, vol points and z-score are in the CSVs below.

Where mismatches concentrate

Strike distance × time of day

• Tradable, all: 17.4% within 100 pts of spot, 45.0% within 300 pts; the busiest band is strikes 300–500 pts above spot (18.3%) and 300–500 below (15.1%).
• Tradable ≥5 pts: closer to the money — 35.1% within 100 pts, 70.6% within 300 pts; the single busiest band is 0–50 pts below spot (14.7%).
• Time of day: the 09:15–09:45 half hour holds 16.5% of big tradable episodes (an even spread would be 8.0%); the last hour holds 10.0% (even: 16%).

Rate view (per 1,000 traded quote-minutes, by moneyness):

Expiry day vs not

• 105 expiry sessions vs 396 other sessions.
• Tradable: 51.5% of episodes happen on expiry day — 19.9 per expiry session vs 5.0 per other session (~4×). Tradable ≥5 pts: 5.5 vs 1.2.
• Liquid ≥5 pts is the opposite: only 6.3% on expiry day; 46.3% sit 4–8 days from expiry (the next-week chain, deep ITM, thinly traded).

Around spot jumps

A "big" 1-minute Nifty move here is ≥11.6 pts (3× the typical 3.9); 9.4% of all minutes are like that. A big 5-minute move is ≥25.5 pts (9.8% of minutes).

Fast spot moves leave some option prices a minute behind; that is where the bigger tradable gaps come from.

Around news

Trading days in the window by news impact (from ./news calendar): none 353, low 57, medium 50, high 39.

High-impact days carry about 2× the tradable mismatches of a quiet day. Scheduled events (mostly RBI policy at ~10:00) show up 3× more than their share of minutes. But the single biggest tradable day had no tagged news, so news explains only part of it.

Top dates

Tradable (all sizes) — the top 10 days hold 34.3% of all 4,055 episodes, spread over 204 days:

Liquid ≥5 pts — busiest days:

CE vs PE, cheap vs rich

• Tradable, all: CE 54.0% / PE 46.0%; cheap 67.2% / rich 32.8%; OTM 77.1%.
• Tradable ≥5 pts: CE 48.4% / PE 51.5%; cheap 48.5% / rich 51.5%; ITM 69.2%. Roughly balanced, so no side is systematically mispriced.
• Liquid ≥5 pts: CE 49.1% / PE 50.9%; cheap 54.3% / rich 45.7%.

Convergence: how long until the gap closed, and who moved

"Half-closed" = the gap fell to half its size or flipped sign; "closed" = down to a quarter (or 0.3 pts). "Market moved to model" = the option price did ≥67% of the closing; "model moved to market" = fair value did (spot or the smile shifted).

Gallery: top 30 tradable mismatches

Each chart shows ±30 min: market price (solid), SVI fair (dashed), spot (grey, right side), dots = minutes the option traded, red line = the flagged minute. Click to enlarge.

Buyer test: does buying "cheap" mismatches make money? — FAIL

Rules (8 variants × 4 hold times × 2 tiers) were scored on train (Sep 2024–Jun 2025). The best on tune (Jul–Dec 2025, needing ≥20 trades) was picked per tier, then run once on the 2026 test. Entry is the minute after the signal, with the costs above. Placebo = random liquid quotes with the same CE/PE, moneyness, time-of-day and price bucket.

Verdict: FAIL for both tiers. Liquid "cheap ≥20 pts, hold 15 min" made money only in the small tune sample (26 trades) and lost ₹39/trade in test (it beat its placebo, but stayed negative). Tradable "cheap ≥20% of fair, hold 1 min" lost ₹182/trade with a 0% win rate in test: the gap (≈0 pts gross) is smaller than costs. The full grid is in buyer_test_all.csv. 8 of 64 2026 test cells show a profit, all in the tradable tier with tiny samples (1–61 trades). With 64 cells, a few positives are what chance alone produces, and none was the rule picked in advance, so they are not evidence of an edge.

Convergence time distribution (all liquid / tradable episodes)

Downloads

top200_tradable_by_pts.csv · top200_tradable_by_pct.csv · top200_tradable_by_vol.csv · top200_tradable_by_z.csv · top200_liquid_by_pts.csv · top200_liquid_by_pct.csv · top200_liquid_by_vol.csv · top200_liquid_by_z.csv · top200_raw_by_pts.csv · top200_raw_by_pct.csv · top200_raw_by_vol.csv · top200_raw_by_z.csv · top20_per_month_tradable_by_pts.csv · top20_per_month_liquid_by_pts.csv · top20_per_month_raw_by_pts.csv · buyer_test_all.csv · buyer_test_selected.csv · clusters.json

Data: Upstox 1-minute Nifty weekly option OHLC/volume/OI (front + next expiry) and Nifty 1-minute spot; news from the in-house India daily news + events calendar. Times are IST. Built 2026-10-18. Research, not advice.