The honest numbers

Who loses in Indian F&O

• FY22-FY24: 93% of 1.13 crore individual F&O traders lost money; aggregate loss ₹1.81 lakh crore (SEBI, Sep 2024).
• FY25: 91% lost; net loss ₹1,05,603 crore; 92% of losses came from options (SEBI, Jul 2025).
• FY26: 87.7% lost; ₹91,685 crore; transaction costs alone turned about 4.4 lakh gross-profitable traders into net losers (SEBI, Aug 2026).

This is why the platform never calls option direction and models option risk only.

What prediction can and cannot do

• Honest out-of-sample daily direction accuracy on liquid stocks: 52-56%. Monthly stock-level R² under 1%.
• Papers claiming 90%+ accuracy are, on inspection, look-ahead leakage, survivorship bias or labels that encode the answer.
• Factor premia (momentum, low volatility, quality) survive in India but roughly halve after publication.
• Professional quant edge comes from breadth, costs and risk control, not from a better single forecast. Medallion's reported hit rate is about 50.75%.

Where retail profit actually comes from

• Staying invested: India's SIP stoppage ratio was 75.6% in FY25. Persistence beats timing.
• Costs: a round trip in cash equities costs about 30-45 bps all-in. Twenty round trips a year is 6-9% of capital.
• Index funds by default: SPIVA India 2025 shows 84% of active large-cap funds underperformed over one year, 92% over three.
• Position sizing and a pre-declared invalidation level matter more than the entry.

How to read Eva

• The stance comes from a ranking engine validated with walk-forward tests, Deflated Sharpe and placebo runs.
• The probability shown is the model's estimate until 30 forward-logged calls have resolved; then it is the realised frequency in that bucket.
• Everything is relative: "outperform" means beating the Nifty 50 median over 20 sessions, not going up.
• Eva's text is generated by a language model from the engine's numbers. It can phrase things well; it cannot know more than the context it was given.