Factors
Code: pipeline/tipsheet/compute/factors.py, publish/factors.py (run inside the indices step).
Bundles: factors/*. Page: /factors/.
Which index stands for each factor
For each factor, the NSE strategy index with the longest live record (earliest launch) among candidates whose history in our data reaches back to the launch, compared with the universe it selects from, so the comparison isolates the factor rule rather than a size difference (choose() in the code):
| Factor | Index | Live since | Compared with |
|---|---|---|---|
| Low volatility | Nifty100 Low Volatility 30 | Jul 2016 | Nifty 100 |
| Quality | Nifty100 Quality 30 (candidate before it: Nifty200 Quality 30, Apr 2018) | Mar 2015 | Nifty 100 |
| Value | Nifty500 Value 50 | Oct 2018 | Nifty 500 |
| Momentum | Nifty200 Momentum 30 | Aug 2020 | Nifty 200 |
| Alpha | Nifty Alpha 50 | Nov 2012 | Nifty 500 (its 300-stock selection universe has no NSE index of its own) |
| Equal weight | Nifty100 Equal Weight | Jun 2013 | Nifty 100 |
Nifty100 Quality 30’s history before June 2026 is filed under its old name (“Nifty100 Qualty30”); compute/indices.load_tri keys rows by NSE index code, so renamed indices keep their history (it had been cut to June 2026 for this and five other indices).
The Nifty500 factor family (Momentum 50, Quality 50, Low Volatility 50) shares one parent but launched in 2024–25, so nearly all of its history is back-calculated; it was not used. Alpha overlaps momentum heavily (correlation of monthly leads 0.77 over five years); it stays because it has the longest live record.
Mid and small caps (second tier), each against its own size universe: Nifty Midcap150 Quality 50 (live Oct 2019) and Midcap150 Momentum 50 (Aug 2022) against the Midcap 150; Smallcap250 Quality 50 (Mar 2023) and Smallcap250 Momentum Quality 100 (Dec 2023) against the Smallcap 250; MidSmallcap400 Momentum Quality 100 (Feb 2024) against the MidSmallcap 400.
Measures
- Lead: the factor’s total return minus its parent’s, in percentage points; beyond a year, the difference of annualised returns. Total-return indices throughout.
- Live and back-calculated: launch dates from NSE factsheets (
compute/trend_launch_dates.csv). The lead per year is computed separately before and after launch. Everything before launch is NSE’s back-calculation, made with the rule already chosen; see C12 indocs/research/site_calculations_spec.mdfor the test across all strategy indices. - Rolling lead: the factor’s 252-session return minus the parent’s, weekly.
- Common window: factor ÷ parent, 100 on the tier’s latest launch date (25 Aug 2020 for the core, 8 Feb 2024 for mid and small caps), so every line is live data over the same market. The table beside it keeps each factor’s lead per year since its own launch. (Rebasing each line at its own launch made the oldest lines, alpha’s from 2012, dominate the scale; with a common start the range is narrow enough for a linear axis.)
- Calendar years: each year’s return minus the parent’s (the latest year to date). Includes back-calculated years; the page labels each factor’s launch year.
- Falls and rallies: average monthly lead in the Nifty 500 TRI’s worst and best tenth of months from the first month most indices cover (May 2005; 26 each), and in the other months. An index with shorter history (Nifty100 Quality 30, from late 2009) is averaged over the tail months it covers, and the page says how many. Descriptive, includes back-calculated years.
- Overlap: correlation of monthly leads over the last five years.
- Risk: volatility and tracking error (standard deviation of daily lead) over the last 756 sessions, annualised; worst and current fall of factor ÷ parent from its running peak.
- Valuation: latest P/E, P/B and dividend yield from NSE’s index valuation data, which for these indices starts on 31 August 2026, so there is no history to rank them against yet.
Limits
- A factor index is one implementation of a factor; another provider’s momentum or quality rule can behave differently.
- Fund returns trail index returns by costs and tracking difference.
- Correlations and falls-and-rallies figures describe the past; with 26 months in each tail they are estimates, not constants.