Data as of 2026-10-09. Outcome pool: Jan 2020 to Sep 2025.
For every company we look at the situation it is in today (valuation, return on capital, margins, debt, growth, 12-month price momentum, drawdown, size and volatility) and find past company-months that looked most like it. We then show what happened to those past situations over the following 12 months, measured against the ApnaFunda All index and against the base rate. It describes history. It does not predict anything.
| Feature | Weight | Non-financial | Financial |
|---|---|---|---|
| P/E | 1 | yes | yes |
| P/S | 1 | yes | — |
| P/B | 1 | — | yes |
| ROCE | 1 | yes | — |
| operating margin | 1 | yes | — |
| ROE | 1 | — | yes |
| debt/equity | 0.75 | yes | — |
| interest cover | 0.5 | yes | — |
| revenue growth | 0.75 | yes | yes |
| 12-month momentum | 1 | yes | yes |
| drawdown from 12-month high | 0.75 | yes | yes |
| size | 0.75 | yes | yes |
| volatility | 0.5 | yes | yes |
Weights were set by judgement before any outcome was looked at and are never tuned on results. Loss-making companies (negative P/E or net profit) are only matched with other loss-making companies. Banks, NBFCs and insurers are compared only with each other and use a different feature set. Every feature is converted, month by month, to its rank among companies in the same pool in that month, so a P/E of 30 in a hot market and a P/E of 22 in a cold one can match if both were equally expensive for their time. Only that month's data is used.
Distance is the weighted root-mean-square gap between the standardised features both companies have. A match needs at least 70% of the query's feature weight, a distance of at most 0.9, and 30 matches at minimum to be shown. We keep up to 50 matches with these exclusions: never the company itself, one match per company (its closest month), and at most 7 from any one calendar month so a single market episode cannot dominate.
The base rate is the distribution of 12-month returns, relative to the market, of every eligible company-month in the same pool, with each month weighted to match the months the analogs came from. It answers: how did a random company do in these same months? Because matches from nearby months share the same market conditions, we also report the number of distinct 12-month windows behind a result, and a bootstrap that resamples months decides whether the analog median differs from the base rate at all. "Not meaningfully different" is the usual outcome and the page copy is written for it.
The walk-forward check has not been run for this build, so skill is shown as not yet measured.