Quantitative research on Indian index options — NIFTY 50, weekly expiries, short premium. I build the data pipeline, test the idea adversarially, and publish the result whether or not it worked. Most of what's below is a null.

M.Tech, AI & ML (BITS Pilani, WILP) · QuantInsti EPAT · LinkedIn

vix-forward-range — Does today's India VIX close predict tomorrow's NIFTY range? Over 1,402 sessions, r = 0.469 and it survives HAC errors, a block bootstrap and a locked holdout — but a five-day-old VIX print scores 0.389 against the same target, so most of it is a slow regime read, not a fresh daily signal. R² = 0.22. Clone and run it: both raw series are committed, no broker account needed.

nifty-short-premium-lab — 71 weeks of real NIFTY options against a retail options-selling playbook. 4 confirmed, 9 null, 3 retracted — including three strategy variants I built myself and then killed. The backwardation-fade spread looked real at p = 0.018 full-sample and was carried entirely by the first eight months.

donchian-option-overlay — A falsification study of the breakout signal every retail options group uses as a timing device. It carries no tradeable edge on NIFTY weeklies. What does survive is narrower: the channel's width forecasts the size of the next move, not its direction. Includes a clean-room re-implementation that imports nothing from the main library, and a deliberate look-ahead probe to prove the pipeline isn't leaking.

nifty-oi-repositioning — Open interest repositioning around intraday structural breaks is strongly associated with break direction, and the association is reactive, not predictive: a real-time signal locked on 2025 parameters does not survive a 2026 out-of-sample test.

greeks-surface-lab — A dependency-free Black-Scholes solver, a Greek-based P&L attribution engine, and a 3D explorer driven by real NIFTY option-chain prints. Every Greek — including vanna, charm and volga — is cross-checked against finite differences of the pricer itself; mypy --strict.

credit-spread-pipeline — A BPS/BCS research pipeline built around one constraint: enforced non-overtrading. One decision per session, hard caps, no re-entries, enforced in risk.py rather than left to discipline. Point-in-time correctness is tested, not assumed.

esco-onet-semantic-alignment — M.Tech dissertation. Embedding-based alignment of the ESCO and O*NET occupational taxonomies, with dominance-gap and entropy diagnostics routing each occupation to accept / review / ambiguous.

Market data used in this research comes from licensed sources and is not redistributed. Where a repository can ship its own data or a synthetic fixture, it does — see each README's reproducibility note.